/[pcre]/code/trunk/pcre_compile.c
ViewVC logotype

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 927 - (show annotations)
Wed Feb 22 15:15:08 2012 UTC (7 years, 7 months ago) by ph10
File MIME type: text/plain
File size: 268808 byte(s)
Error occurred while calculating annotation data.
Correct and tidy up comments relating to OP_NOT (no code changes).
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80
81 /*************************************************
82 * Code parameters and static tables *
83 *************************************************/
84
85 /* This value specifies the size of stack workspace that is used during the
86 first pre-compile phase that determines how much memory is required. The regex
87 is partly compiled into this space, but the compiled parts are discarded as
88 soon as they can be, so that hopefully there will never be an overrun. The code
89 does, however, check for an overrun. The largest amount I've seen used is 218,
90 so this number is very generous.
91
92 The same workspace is used during the second, actual compile phase for
93 remembering forward references to groups so that they can be filled in at the
94 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
95 is 4 there is plenty of room for most patterns. However, the memory can get
96 filled up by repetitions of forward references, for example patterns like
97 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
98 that the workspace is expanded using malloc() in this situation. The value
99 below is therefore a minimum, and we put a maximum on it for safety. The
100 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
101 kicks in at the same number of forward references in all cases. */
102
103 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
104 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
105
106 /* The overrun tests check for a slightly smaller size so that they detect the
107 overrun before it actually does run off the end of the data block. */
108
109 #define WORK_SIZE_SAFETY_MARGIN (100)
110
111 /* Private flags added to firstchar and reqchar. */
112
113 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
114 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
115
116 /* Repeated character flags. */
117
118 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
119
120 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
121 are simple data values; negative values are for special things like \d and so
122 on. Zero means further processing is needed (for things like \x), or the escape
123 is invalid. */
124
125 #ifndef EBCDIC
126
127 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
128 in UTF-8 mode. */
129
130 static const short int escapes[] = {
131 0, 0,
132 0, 0,
133 0, 0,
134 0, 0,
135 0, 0,
136 CHAR_COLON, CHAR_SEMICOLON,
137 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
138 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
139 CHAR_COMMERCIAL_AT, -ESC_A,
140 -ESC_B, -ESC_C,
141 -ESC_D, -ESC_E,
142 0, -ESC_G,
143 -ESC_H, 0,
144 0, -ESC_K,
145 0, 0,
146 -ESC_N, 0,
147 -ESC_P, -ESC_Q,
148 -ESC_R, -ESC_S,
149 0, 0,
150 -ESC_V, -ESC_W,
151 -ESC_X, 0,
152 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
153 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
154 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
155 CHAR_GRAVE_ACCENT, 7,
156 -ESC_b, 0,
157 -ESC_d, ESC_e,
158 ESC_f, 0,
159 -ESC_h, 0,
160 0, -ESC_k,
161 0, 0,
162 ESC_n, 0,
163 -ESC_p, 0,
164 ESC_r, -ESC_s,
165 ESC_tee, 0,
166 -ESC_v, -ESC_w,
167 0, 0,
168 -ESC_z
169 };
170
171 #else
172
173 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
174
175 static const short int escapes[] = {
176 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
177 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
178 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
179 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
180 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
181 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
182 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
183 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
184 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
185 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
186 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
187 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
188 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
189 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
190 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
191 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
192 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
193 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
194 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
195 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
196 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
197 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
198 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
199 };
200 #endif
201
202
203 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
204 searched linearly. Put all the names into a single string, in order to reduce
205 the number of relocations when a shared library is dynamically linked. The
206 string is built from string macros so that it works in UTF-8 mode on EBCDIC
207 platforms. */
208
209 typedef struct verbitem {
210 int len; /* Length of verb name */
211 int op; /* Op when no arg, or -1 if arg mandatory */
212 int op_arg; /* Op when arg present, or -1 if not allowed */
213 } verbitem;
214
215 static const char verbnames[] =
216 "\0" /* Empty name is a shorthand for MARK */
217 STRING_MARK0
218 STRING_ACCEPT0
219 STRING_COMMIT0
220 STRING_F0
221 STRING_FAIL0
222 STRING_PRUNE0
223 STRING_SKIP0
224 STRING_THEN;
225
226 static const verbitem verbs[] = {
227 { 0, -1, OP_MARK },
228 { 4, -1, OP_MARK },
229 { 6, OP_ACCEPT, -1 },
230 { 6, OP_COMMIT, -1 },
231 { 1, OP_FAIL, -1 },
232 { 4, OP_FAIL, -1 },
233 { 5, OP_PRUNE, OP_PRUNE_ARG },
234 { 4, OP_SKIP, OP_SKIP_ARG },
235 { 4, OP_THEN, OP_THEN_ARG }
236 };
237
238 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
239
240
241 /* Tables of names of POSIX character classes and their lengths. The names are
242 now all in a single string, to reduce the number of relocations when a shared
243 library is dynamically loaded. The list of lengths is terminated by a zero
244 length entry. The first three must be alpha, lower, upper, as this is assumed
245 for handling case independence. */
246
247 static const char posix_names[] =
248 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
249 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
250 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
251 STRING_word0 STRING_xdigit;
252
253 static const pcre_uint8 posix_name_lengths[] = {
254 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
255
256 /* Table of class bit maps for each POSIX class. Each class is formed from a
257 base map, with an optional addition or removal of another map. Then, for some
258 classes, there is some additional tweaking: for [:blank:] the vertical space
259 characters are removed, and for [:alpha:] and [:alnum:] the underscore
260 character is removed. The triples in the table consist of the base map offset,
261 second map offset or -1 if no second map, and a non-negative value for map
262 addition or a negative value for map subtraction (if there are two maps). The
263 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
264 remove vertical space characters, 2 => remove underscore. */
265
266 static const int posix_class_maps[] = {
267 cbit_word, cbit_digit, -2, /* alpha */
268 cbit_lower, -1, 0, /* lower */
269 cbit_upper, -1, 0, /* upper */
270 cbit_word, -1, 2, /* alnum - word without underscore */
271 cbit_print, cbit_cntrl, 0, /* ascii */
272 cbit_space, -1, 1, /* blank - a GNU extension */
273 cbit_cntrl, -1, 0, /* cntrl */
274 cbit_digit, -1, 0, /* digit */
275 cbit_graph, -1, 0, /* graph */
276 cbit_print, -1, 0, /* print */
277 cbit_punct, -1, 0, /* punct */
278 cbit_space, -1, 0, /* space */
279 cbit_word, -1, 0, /* word - a Perl extension */
280 cbit_xdigit,-1, 0 /* xdigit */
281 };
282
283 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
284 substitutes must be in the order of the names, defined above, and there are
285 both positive and negative cases. NULL means no substitute. */
286
287 #ifdef SUPPORT_UCP
288 static const pcre_uchar string_PNd[] = {
289 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
290 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
291 static const pcre_uchar string_pNd[] = {
292 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
293 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
294 static const pcre_uchar string_PXsp[] = {
295 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
296 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
297 static const pcre_uchar string_pXsp[] = {
298 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
299 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
300 static const pcre_uchar string_PXwd[] = {
301 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303 static const pcre_uchar string_pXwd[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306
307 static const pcre_uchar *substitutes[] = {
308 string_PNd, /* \D */
309 string_pNd, /* \d */
310 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
311 string_pXsp, /* \s */
312 string_PXwd, /* \W */
313 string_pXwd /* \w */
314 };
315
316 static const pcre_uchar string_pL[] = {
317 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
318 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
319 static const pcre_uchar string_pLl[] = {
320 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
321 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
322 static const pcre_uchar string_pLu[] = {
323 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
324 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
325 static const pcre_uchar string_pXan[] = {
326 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
327 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
328 static const pcre_uchar string_h[] = {
329 CHAR_BACKSLASH, CHAR_h, '\0' };
330 static const pcre_uchar string_pXps[] = {
331 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_PL[] = {
334 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_PLl[] = {
337 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_PLu[] = {
340 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
341 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
342 static const pcre_uchar string_PXan[] = {
343 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
344 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
345 static const pcre_uchar string_H[] = {
346 CHAR_BACKSLASH, CHAR_H, '\0' };
347 static const pcre_uchar string_PXps[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350
351 static const pcre_uchar *posix_substitutes[] = {
352 string_pL, /* alpha */
353 string_pLl, /* lower */
354 string_pLu, /* upper */
355 string_pXan, /* alnum */
356 NULL, /* ascii */
357 string_h, /* blank */
358 NULL, /* cntrl */
359 string_pNd, /* digit */
360 NULL, /* graph */
361 NULL, /* print */
362 NULL, /* punct */
363 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
364 string_pXwd, /* word */
365 NULL, /* xdigit */
366 /* Negated cases */
367 string_PL, /* ^alpha */
368 string_PLl, /* ^lower */
369 string_PLu, /* ^upper */
370 string_PXan, /* ^alnum */
371 NULL, /* ^ascii */
372 string_H, /* ^blank */
373 NULL, /* ^cntrl */
374 string_PNd, /* ^digit */
375 NULL, /* ^graph */
376 NULL, /* ^print */
377 NULL, /* ^punct */
378 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
379 string_PXwd, /* ^word */
380 NULL /* ^xdigit */
381 };
382 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
383 #endif
384
385 #define STRING(a) # a
386 #define XSTRING(s) STRING(s)
387
388 /* The texts of compile-time error messages. These are "char *" because they
389 are passed to the outside world. Do not ever re-use any error number, because
390 they are documented. Always add a new error instead. Messages marked DEAD below
391 are no longer used. This used to be a table of strings, but in order to reduce
392 the number of relocations needed when a shared library is loaded dynamically,
393 it is now one long string. We cannot use a table of offsets, because the
394 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
395 simply count through to the one we want - this isn't a performance issue
396 because these strings are used only when there is a compilation error.
397
398 Each substring ends with \0 to insert a null character. This includes the final
399 substring, so that the whole string ends with \0\0, which can be detected when
400 counting through. */
401
402 static const char error_texts[] =
403 "no error\0"
404 "\\ at end of pattern\0"
405 "\\c at end of pattern\0"
406 "unrecognized character follows \\\0"
407 "numbers out of order in {} quantifier\0"
408 /* 5 */
409 "number too big in {} quantifier\0"
410 "missing terminating ] for character class\0"
411 "invalid escape sequence in character class\0"
412 "range out of order in character class\0"
413 "nothing to repeat\0"
414 /* 10 */
415 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
416 "internal error: unexpected repeat\0"
417 "unrecognized character after (? or (?-\0"
418 "POSIX named classes are supported only within a class\0"
419 "missing )\0"
420 /* 15 */
421 "reference to non-existent subpattern\0"
422 "erroffset passed as NULL\0"
423 "unknown option bit(s) set\0"
424 "missing ) after comment\0"
425 "parentheses nested too deeply\0" /** DEAD **/
426 /* 20 */
427 "regular expression is too large\0"
428 "failed to get memory\0"
429 "unmatched parentheses\0"
430 "internal error: code overflow\0"
431 "unrecognized character after (?<\0"
432 /* 25 */
433 "lookbehind assertion is not fixed length\0"
434 "malformed number or name after (?(\0"
435 "conditional group contains more than two branches\0"
436 "assertion expected after (?(\0"
437 "(?R or (?[+-]digits must be followed by )\0"
438 /* 30 */
439 "unknown POSIX class name\0"
440 "POSIX collating elements are not supported\0"
441 "this version of PCRE is compiled without UTF support\0"
442 "spare error\0" /** DEAD **/
443 "character value in \\x{...} sequence is too large\0"
444 /* 35 */
445 "invalid condition (?(0)\0"
446 "\\C not allowed in lookbehind assertion\0"
447 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
448 "number after (?C is > 255\0"
449 "closing ) for (?C expected\0"
450 /* 40 */
451 "recursive call could loop indefinitely\0"
452 "unrecognized character after (?P\0"
453 "syntax error in subpattern name (missing terminator)\0"
454 "two named subpatterns have the same name\0"
455 "invalid UTF-8 string\0"
456 /* 45 */
457 "support for \\P, \\p, and \\X has not been compiled\0"
458 "malformed \\P or \\p sequence\0"
459 "unknown property name after \\P or \\p\0"
460 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
461 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
462 /* 50 */
463 "repeated subpattern is too long\0" /** DEAD **/
464 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
465 "internal error: overran compiling workspace\0"
466 "internal error: previously-checked referenced subpattern not found\0"
467 "DEFINE group contains more than one branch\0"
468 /* 55 */
469 "repeating a DEFINE group is not allowed\0" /** DEAD **/
470 "inconsistent NEWLINE options\0"
471 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
472 "a numbered reference must not be zero\0"
473 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
474 /* 60 */
475 "(*VERB) not recognized\0"
476 "number is too big\0"
477 "subpattern name expected\0"
478 "digit expected after (?+\0"
479 "] is an invalid data character in JavaScript compatibility mode\0"
480 /* 65 */
481 "different names for subpatterns of the same number are not allowed\0"
482 "(*MARK) must have an argument\0"
483 "this version of PCRE is not compiled with Unicode property support\0"
484 "\\c must be followed by an ASCII character\0"
485 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
486 /* 70 */
487 "internal error: unknown opcode in find_fixedlength()\0"
488 "\\N is not supported in a class\0"
489 "too many forward references\0"
490 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
491 "invalid UTF-16 string\0"
492 ;
493
494 /* Table to identify digits and hex digits. This is used when compiling
495 patterns. Note that the tables in chartables are dependent on the locale, and
496 may mark arbitrary characters as digits - but the PCRE compiling code expects
497 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
498 a private table here. It costs 256 bytes, but it is a lot faster than doing
499 character value tests (at least in some simple cases I timed), and in some
500 applications one wants PCRE to compile efficiently as well as match
501 efficiently.
502
503 For convenience, we use the same bit definitions as in chartables:
504
505 0x04 decimal digit
506 0x08 hexadecimal digit
507
508 Then we can use ctype_digit and ctype_xdigit in the code. */
509
510 /* Using a simple comparison for decimal numbers rather than a memory read
511 is much faster, and the resulting code is simpler (the compiler turns it
512 into a subtraction and unsigned comparison). */
513
514 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
515
516 #ifndef EBCDIC
517
518 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
519 UTF-8 mode. */
520
521 static const pcre_uint8 digitab[] =
522 {
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
529 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
530 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
531 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
535 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
555
556 #else
557
558 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
559
560 static const pcre_uint8 digitab[] =
561 {
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
578 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
586 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
592 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
593 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
594
595 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
596 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
597 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
598 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
600 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
604 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
605 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
607 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
609 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
612 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
613 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
614 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
615 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
616 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
617 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
618 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
619 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
620 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
621 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
622 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
623 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
624 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
625 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
626 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
627 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
628 #endif
629
630
631 /* Definition to allow mutual recursion */
632
633 static BOOL
634 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
635 int *, int *, branch_chain *, compile_data *, int *);
636
637
638
639 /*************************************************
640 * Find an error text *
641 *************************************************/
642
643 /* The error texts are now all in one long string, to save on relocations. As
644 some of the text is of unknown length, we can't use a table of offsets.
645 Instead, just count through the strings. This is not a performance issue
646 because it happens only when there has been a compilation error.
647
648 Argument: the error number
649 Returns: pointer to the error string
650 */
651
652 static const char *
653 find_error_text(int n)
654 {
655 const char *s = error_texts;
656 for (; n > 0; n--)
657 {
658 while (*s++ != 0) {};
659 if (*s == 0) return "Error text not found (please report)";
660 }
661 return s;
662 }
663
664
665 /*************************************************
666 * Expand the workspace *
667 *************************************************/
668
669 /* This function is called during the second compiling phase, if the number of
670 forward references fills the existing workspace, which is originally a block on
671 the stack. A larger block is obtained from malloc() unless the ultimate limit
672 has been reached or the increase will be rather small.
673
674 Argument: pointer to the compile data block
675 Returns: 0 if all went well, else an error number
676 */
677
678 static int
679 expand_workspace(compile_data *cd)
680 {
681 pcre_uchar *newspace;
682 int newsize = cd->workspace_size * 2;
683
684 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
685 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
686 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
687 return ERR72;
688
689 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
690 if (newspace == NULL) return ERR21;
691 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
692 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
693 if (cd->workspace_size > COMPILE_WORK_SIZE)
694 (PUBL(free))((void *)cd->start_workspace);
695 cd->start_workspace = newspace;
696 cd->workspace_size = newsize;
697 return 0;
698 }
699
700
701
702 /*************************************************
703 * Check for counted repeat *
704 *************************************************/
705
706 /* This function is called when a '{' is encountered in a place where it might
707 start a quantifier. It looks ahead to see if it really is a quantifier or not.
708 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
709 where the ddds are digits.
710
711 Arguments:
712 p pointer to the first char after '{'
713
714 Returns: TRUE or FALSE
715 */
716
717 static BOOL
718 is_counted_repeat(const pcre_uchar *p)
719 {
720 if (!IS_DIGIT(*p)) return FALSE;
721 p++;
722 while (IS_DIGIT(*p)) p++;
723 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
724
725 if (*p++ != CHAR_COMMA) return FALSE;
726 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
727
728 if (!IS_DIGIT(*p)) return FALSE;
729 p++;
730 while (IS_DIGIT(*p)) p++;
731
732 return (*p == CHAR_RIGHT_CURLY_BRACKET);
733 }
734
735
736
737 /*************************************************
738 * Handle escapes *
739 *************************************************/
740
741 /* This function is called when a \ has been encountered. It either returns a
742 positive value for a simple escape such as \n, or a negative value which
743 encodes one of the more complicated things such as \d. A backreference to group
744 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
745 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
746 ptr is pointing at the \. On exit, it is on the final character of the escape
747 sequence.
748
749 Arguments:
750 ptrptr points to the pattern position pointer
751 errorcodeptr points to the errorcode variable
752 bracount number of previous extracting brackets
753 options the options bits
754 isclass TRUE if inside a character class
755
756 Returns: zero or positive => a data character
757 negative => a special escape sequence
758 on error, errorcodeptr is set
759 */
760
761 static int
762 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
763 int options, BOOL isclass)
764 {
765 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
766 BOOL utf = (options & PCRE_UTF8) != 0;
767 const pcre_uchar *ptr = *ptrptr + 1;
768 pcre_int32 c;
769 int i;
770
771 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
772 ptr--; /* Set pointer back to the last byte */
773
774 /* If backslash is at the end of the pattern, it's an error. */
775
776 if (c == 0) *errorcodeptr = ERR1;
777
778 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
779 in a table. A non-zero result is something that can be returned immediately.
780 Otherwise further processing may be required. */
781
782 #ifndef EBCDIC /* ASCII/UTF-8 coding */
783 /* Not alphanumeric */
784 else if (c < CHAR_0 || c > CHAR_z) {}
785 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
786
787 #else /* EBCDIC coding */
788 /* Not alphanumeric */
789 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
790 else if ((i = escapes[c - 0x48]) != 0) c = i;
791 #endif
792
793 /* Escapes that need further processing, or are illegal. */
794
795 else
796 {
797 const pcre_uchar *oldptr;
798 BOOL braced, negated;
799
800 switch (c)
801 {
802 /* A number of Perl escapes are not handled by PCRE. We give an explicit
803 error. */
804
805 case CHAR_l:
806 case CHAR_L:
807 *errorcodeptr = ERR37;
808 break;
809
810 case CHAR_u:
811 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
812 {
813 /* In JavaScript, \u must be followed by four hexadecimal numbers.
814 Otherwise it is a lowercase u letter. */
815 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
816 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
817 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
818 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
819 {
820 c = 0;
821 for (i = 0; i < 4; ++i)
822 {
823 register int cc = *(++ptr);
824 #ifndef EBCDIC /* ASCII/UTF-8 coding */
825 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
826 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
827 #else /* EBCDIC coding */
828 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
829 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
830 #endif
831 }
832 }
833 }
834 else
835 *errorcodeptr = ERR37;
836 break;
837
838 case CHAR_U:
839 /* In JavaScript, \U is an uppercase U letter. */
840 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
841 break;
842
843 /* In a character class, \g is just a literal "g". Outside a character
844 class, \g must be followed by one of a number of specific things:
845
846 (1) A number, either plain or braced. If positive, it is an absolute
847 backreference. If negative, it is a relative backreference. This is a Perl
848 5.10 feature.
849
850 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
851 is part of Perl's movement towards a unified syntax for back references. As
852 this is synonymous with \k{name}, we fudge it up by pretending it really
853 was \k.
854
855 (3) For Oniguruma compatibility we also support \g followed by a name or a
856 number either in angle brackets or in single quotes. However, these are
857 (possibly recursive) subroutine calls, _not_ backreferences. Just return
858 the -ESC_g code (cf \k). */
859
860 case CHAR_g:
861 if (isclass) break;
862 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
863 {
864 c = -ESC_g;
865 break;
866 }
867
868 /* Handle the Perl-compatible cases */
869
870 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
871 {
872 const pcre_uchar *p;
873 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
874 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
875 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
876 {
877 c = -ESC_k;
878 break;
879 }
880 braced = TRUE;
881 ptr++;
882 }
883 else braced = FALSE;
884
885 if (ptr[1] == CHAR_MINUS)
886 {
887 negated = TRUE;
888 ptr++;
889 }
890 else negated = FALSE;
891
892 /* The integer range is limited by the machine's int representation. */
893 c = 0;
894 while (IS_DIGIT(ptr[1]))
895 {
896 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
897 {
898 c = -1;
899 break;
900 }
901 c = c * 10 + *(++ptr) - CHAR_0;
902 }
903 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
904 {
905 while (IS_DIGIT(ptr[1]))
906 ptr++;
907 *errorcodeptr = ERR61;
908 break;
909 }
910
911 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
912 {
913 *errorcodeptr = ERR57;
914 break;
915 }
916
917 if (c == 0)
918 {
919 *errorcodeptr = ERR58;
920 break;
921 }
922
923 if (negated)
924 {
925 if (c > bracount)
926 {
927 *errorcodeptr = ERR15;
928 break;
929 }
930 c = bracount - (c - 1);
931 }
932
933 c = -(ESC_REF + c);
934 break;
935
936 /* The handling of escape sequences consisting of a string of digits
937 starting with one that is not zero is not straightforward. By experiment,
938 the way Perl works seems to be as follows:
939
940 Outside a character class, the digits are read as a decimal number. If the
941 number is less than 10, or if there are that many previous extracting
942 left brackets, then it is a back reference. Otherwise, up to three octal
943 digits are read to form an escaped byte. Thus \123 is likely to be octal
944 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
945 value is greater than 377, the least significant 8 bits are taken. Inside a
946 character class, \ followed by a digit is always an octal number. */
947
948 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
949 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
950
951 if (!isclass)
952 {
953 oldptr = ptr;
954 /* The integer range is limited by the machine's int representation. */
955 c -= CHAR_0;
956 while (IS_DIGIT(ptr[1]))
957 {
958 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
959 {
960 c = -1;
961 break;
962 }
963 c = c * 10 + *(++ptr) - CHAR_0;
964 }
965 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
966 {
967 while (IS_DIGIT(ptr[1]))
968 ptr++;
969 *errorcodeptr = ERR61;
970 break;
971 }
972 if (c < 10 || c <= bracount)
973 {
974 c = -(ESC_REF + c);
975 break;
976 }
977 ptr = oldptr; /* Put the pointer back and fall through */
978 }
979
980 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
981 generates a binary zero byte and treats the digit as a following literal.
982 Thus we have to pull back the pointer by one. */
983
984 if ((c = *ptr) >= CHAR_8)
985 {
986 ptr--;
987 c = 0;
988 break;
989 }
990
991 /* \0 always starts an octal number, but we may drop through to here with a
992 larger first octal digit. The original code used just to take the least
993 significant 8 bits of octal numbers (I think this is what early Perls used
994 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
995 but no more than 3 octal digits. */
996
997 case CHAR_0:
998 c -= CHAR_0;
999 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1000 c = c * 8 + *(++ptr) - CHAR_0;
1001 #ifdef COMPILE_PCRE8
1002 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1003 #endif
1004 break;
1005
1006 /* \x is complicated. \x{ddd} is a character number which can be greater
1007 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1008 If not, { is treated as a data character. */
1009
1010 case CHAR_x:
1011 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1012 {
1013 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1014 Otherwise it is a lowercase x letter. */
1015 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1016 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1017 {
1018 c = 0;
1019 for (i = 0; i < 2; ++i)
1020 {
1021 register int cc = *(++ptr);
1022 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1023 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1024 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1025 #else /* EBCDIC coding */
1026 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1027 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1028 #endif
1029 }
1030 }
1031 break;
1032 }
1033
1034 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1035 {
1036 const pcre_uchar *pt = ptr + 2;
1037
1038 c = 0;
1039 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1040 {
1041 register int cc = *pt++;
1042 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1043
1044 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1045 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1046 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1047 #else /* EBCDIC coding */
1048 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1049 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1050 #endif
1051
1052 #ifdef COMPILE_PCRE8
1053 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1054 #else
1055 #ifdef COMPILE_PCRE16
1056 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1057 #endif
1058 #endif
1059 }
1060
1061 if (c < 0)
1062 {
1063 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1064 *errorcodeptr = ERR34;
1065 }
1066
1067 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1068 {
1069 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1070 ptr = pt;
1071 break;
1072 }
1073
1074 /* If the sequence of hex digits does not end with '}', then we don't
1075 recognize this construct; fall through to the normal \x handling. */
1076 }
1077
1078 /* Read just a single-byte hex-defined char */
1079
1080 c = 0;
1081 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1082 {
1083 int cc; /* Some compilers don't like */
1084 cc = *(++ptr); /* ++ in initializers */
1085 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1086 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1087 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1088 #else /* EBCDIC coding */
1089 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1090 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1091 #endif
1092 }
1093 break;
1094
1095 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1096 An error is given if the byte following \c is not an ASCII character. This
1097 coding is ASCII-specific, but then the whole concept of \cx is
1098 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1099
1100 case CHAR_c:
1101 c = *(++ptr);
1102 if (c == 0)
1103 {
1104 *errorcodeptr = ERR2;
1105 break;
1106 }
1107 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1108 if (c > 127) /* Excludes all non-ASCII in either mode */
1109 {
1110 *errorcodeptr = ERR68;
1111 break;
1112 }
1113 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1114 c ^= 0x40;
1115 #else /* EBCDIC coding */
1116 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1117 c ^= 0xC0;
1118 #endif
1119 break;
1120
1121 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1122 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1123 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1124 odd, but there used to be some cases other than the default, and there may
1125 be again in future, so I haven't "optimized" it. */
1126
1127 default:
1128 if ((options & PCRE_EXTRA) != 0) switch(c)
1129 {
1130 default:
1131 *errorcodeptr = ERR3;
1132 break;
1133 }
1134 break;
1135 }
1136 }
1137
1138 /* Perl supports \N{name} for character names, as well as plain \N for "not
1139 newline". PCRE does not support \N{name}. However, it does support
1140 quantification such as \N{2,3}. */
1141
1142 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1143 !is_counted_repeat(ptr+2))
1144 *errorcodeptr = ERR37;
1145
1146 /* If PCRE_UCP is set, we change the values for \d etc. */
1147
1148 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1149 c -= (ESC_DU - ESC_D);
1150
1151 /* Set the pointer to the final character before returning. */
1152
1153 *ptrptr = ptr;
1154 return c;
1155 }
1156
1157
1158
1159 #ifdef SUPPORT_UCP
1160 /*************************************************
1161 * Handle \P and \p *
1162 *************************************************/
1163
1164 /* This function is called after \P or \p has been encountered, provided that
1165 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1166 pointing at the P or p. On exit, it is pointing at the final character of the
1167 escape sequence.
1168
1169 Argument:
1170 ptrptr points to the pattern position pointer
1171 negptr points to a boolean that is set TRUE for negation else FALSE
1172 dptr points to an int that is set to the detailed property value
1173 errorcodeptr points to the error code variable
1174
1175 Returns: type value from ucp_type_table, or -1 for an invalid type
1176 */
1177
1178 static int
1179 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1180 {
1181 int c, i, bot, top;
1182 const pcre_uchar *ptr = *ptrptr;
1183 pcre_uchar name[32];
1184
1185 c = *(++ptr);
1186 if (c == 0) goto ERROR_RETURN;
1187
1188 *negptr = FALSE;
1189
1190 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1191 negation. */
1192
1193 if (c == CHAR_LEFT_CURLY_BRACKET)
1194 {
1195 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1196 {
1197 *negptr = TRUE;
1198 ptr++;
1199 }
1200 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1201 {
1202 c = *(++ptr);
1203 if (c == 0) goto ERROR_RETURN;
1204 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1205 name[i] = c;
1206 }
1207 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1208 name[i] = 0;
1209 }
1210
1211 /* Otherwise there is just one following character */
1212
1213 else
1214 {
1215 name[0] = c;
1216 name[1] = 0;
1217 }
1218
1219 *ptrptr = ptr;
1220
1221 /* Search for a recognized property name using binary chop */
1222
1223 bot = 0;
1224 top = PRIV(utt_size);
1225
1226 while (bot < top)
1227 {
1228 i = (bot + top) >> 1;
1229 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1230 if (c == 0)
1231 {
1232 *dptr = PRIV(utt)[i].value;
1233 return PRIV(utt)[i].type;
1234 }
1235 if (c > 0) bot = i + 1; else top = i;
1236 }
1237
1238 *errorcodeptr = ERR47;
1239 *ptrptr = ptr;
1240 return -1;
1241
1242 ERROR_RETURN:
1243 *errorcodeptr = ERR46;
1244 *ptrptr = ptr;
1245 return -1;
1246 }
1247 #endif
1248
1249
1250
1251
1252 /*************************************************
1253 * Read repeat counts *
1254 *************************************************/
1255
1256 /* Read an item of the form {n,m} and return the values. This is called only
1257 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1258 so the syntax is guaranteed to be correct, but we need to check the values.
1259
1260 Arguments:
1261 p pointer to first char after '{'
1262 minp pointer to int for min
1263 maxp pointer to int for max
1264 returned as -1 if no max
1265 errorcodeptr points to error code variable
1266
1267 Returns: pointer to '}' on success;
1268 current ptr on error, with errorcodeptr set non-zero
1269 */
1270
1271 static const pcre_uchar *
1272 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1273 {
1274 int min = 0;
1275 int max = -1;
1276
1277 /* Read the minimum value and do a paranoid check: a negative value indicates
1278 an integer overflow. */
1279
1280 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1281 if (min < 0 || min > 65535)
1282 {
1283 *errorcodeptr = ERR5;
1284 return p;
1285 }
1286
1287 /* Read the maximum value if there is one, and again do a paranoid on its size.
1288 Also, max must not be less than min. */
1289
1290 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1291 {
1292 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1293 {
1294 max = 0;
1295 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1296 if (max < 0 || max > 65535)
1297 {
1298 *errorcodeptr = ERR5;
1299 return p;
1300 }
1301 if (max < min)
1302 {
1303 *errorcodeptr = ERR4;
1304 return p;
1305 }
1306 }
1307 }
1308
1309 /* Fill in the required variables, and pass back the pointer to the terminating
1310 '}'. */
1311
1312 *minp = min;
1313 *maxp = max;
1314 return p;
1315 }
1316
1317
1318
1319 /*************************************************
1320 * Subroutine for finding forward reference *
1321 *************************************************/
1322
1323 /* This recursive function is called only from find_parens() below. The
1324 top-level call starts at the beginning of the pattern. All other calls must
1325 start at a parenthesis. It scans along a pattern's text looking for capturing
1326 subpatterns, and counting them. If it finds a named pattern that matches the
1327 name it is given, it returns its number. Alternatively, if the name is NULL, it
1328 returns when it reaches a given numbered subpattern. Recursion is used to keep
1329 track of subpatterns that reset the capturing group numbers - the (?| feature.
1330
1331 This function was originally called only from the second pass, in which we know
1332 that if (?< or (?' or (?P< is encountered, the name will be correctly
1333 terminated because that is checked in the first pass. There is now one call to
1334 this function in the first pass, to check for a recursive back reference by
1335 name (so that we can make the whole group atomic). In this case, we need check
1336 only up to the current position in the pattern, and that is still OK because
1337 and previous occurrences will have been checked. To make this work, the test
1338 for "end of pattern" is a check against cd->end_pattern in the main loop,
1339 instead of looking for a binary zero. This means that the special first-pass
1340 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1341 processing items within the loop are OK, because afterwards the main loop will
1342 terminate.)
1343
1344 Arguments:
1345 ptrptr address of the current character pointer (updated)
1346 cd compile background data
1347 name name to seek, or NULL if seeking a numbered subpattern
1348 lorn name length, or subpattern number if name is NULL
1349 xmode TRUE if we are in /x mode
1350 utf TRUE if we are in UTF-8 / UTF-16 mode
1351 count pointer to the current capturing subpattern number (updated)
1352
1353 Returns: the number of the named subpattern, or -1 if not found
1354 */
1355
1356 static int
1357 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1358 BOOL xmode, BOOL utf, int *count)
1359 {
1360 pcre_uchar *ptr = *ptrptr;
1361 int start_count = *count;
1362 int hwm_count = start_count;
1363 BOOL dup_parens = FALSE;
1364
1365 /* If the first character is a parenthesis, check on the type of group we are
1366 dealing with. The very first call may not start with a parenthesis. */
1367
1368 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1369 {
1370 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1371
1372 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1373
1374 /* Handle a normal, unnamed capturing parenthesis. */
1375
1376 else if (ptr[1] != CHAR_QUESTION_MARK)
1377 {
1378 *count += 1;
1379 if (name == NULL && *count == lorn) return *count;
1380 ptr++;
1381 }
1382
1383 /* All cases now have (? at the start. Remember when we are in a group
1384 where the parenthesis numbers are duplicated. */
1385
1386 else if (ptr[2] == CHAR_VERTICAL_LINE)
1387 {
1388 ptr += 3;
1389 dup_parens = TRUE;
1390 }
1391
1392 /* Handle comments; all characters are allowed until a ket is reached. */
1393
1394 else if (ptr[2] == CHAR_NUMBER_SIGN)
1395 {
1396 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1397 goto FAIL_EXIT;
1398 }
1399
1400 /* Handle a condition. If it is an assertion, just carry on so that it
1401 is processed as normal. If not, skip to the closing parenthesis of the
1402 condition (there can't be any nested parens). */
1403
1404 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1405 {
1406 ptr += 2;
1407 if (ptr[1] != CHAR_QUESTION_MARK)
1408 {
1409 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1410 if (*ptr != 0) ptr++;
1411 }
1412 }
1413
1414 /* Start with (? but not a condition. */
1415
1416 else
1417 {
1418 ptr += 2;
1419 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1420
1421 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1422
1423 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1424 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1425 {
1426 int term;
1427 const pcre_uchar *thisname;
1428 *count += 1;
1429 if (name == NULL && *count == lorn) return *count;
1430 term = *ptr++;
1431 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1432 thisname = ptr;
1433 while (*ptr != term) ptr++;
1434 if (name != NULL && lorn == ptr - thisname &&
1435 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1436 return *count;
1437 term++;
1438 }
1439 }
1440 }
1441
1442 /* Past any initial parenthesis handling, scan for parentheses or vertical
1443 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1444 first-pass call when this value is temporarily adjusted to stop at the current
1445 position. So DO NOT change this to a test for binary zero. */
1446
1447 for (; ptr < cd->end_pattern; ptr++)
1448 {
1449 /* Skip over backslashed characters and also entire \Q...\E */
1450
1451 if (*ptr == CHAR_BACKSLASH)
1452 {
1453 if (*(++ptr) == 0) goto FAIL_EXIT;
1454 if (*ptr == CHAR_Q) for (;;)
1455 {
1456 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1457 if (*ptr == 0) goto FAIL_EXIT;
1458 if (*(++ptr) == CHAR_E) break;
1459 }
1460 continue;
1461 }
1462
1463 /* Skip over character classes; this logic must be similar to the way they
1464 are handled for real. If the first character is '^', skip it. Also, if the
1465 first few characters (either before or after ^) are \Q\E or \E we skip them
1466 too. This makes for compatibility with Perl. Note the use of STR macros to
1467 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1468
1469 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1470 {
1471 BOOL negate_class = FALSE;
1472 for (;;)
1473 {
1474 if (ptr[1] == CHAR_BACKSLASH)
1475 {
1476 if (ptr[2] == CHAR_E)
1477 ptr+= 2;
1478 else if (STRNCMP_UC_C8(ptr + 2,
1479 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1480 ptr += 4;
1481 else
1482 break;
1483 }
1484 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1485 {
1486 negate_class = TRUE;
1487 ptr++;
1488 }
1489 else break;
1490 }
1491
1492 /* If the next character is ']', it is a data character that must be
1493 skipped, except in JavaScript compatibility mode. */
1494
1495 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1496 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1497 ptr++;
1498
1499 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1500 {
1501 if (*ptr == 0) return -1;
1502 if (*ptr == CHAR_BACKSLASH)
1503 {
1504 if (*(++ptr) == 0) goto FAIL_EXIT;
1505 if (*ptr == CHAR_Q) for (;;)
1506 {
1507 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1508 if (*ptr == 0) goto FAIL_EXIT;
1509 if (*(++ptr) == CHAR_E) break;
1510 }
1511 continue;
1512 }
1513 }
1514 continue;
1515 }
1516
1517 /* Skip comments in /x mode */
1518
1519 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1520 {
1521 ptr++;
1522 while (*ptr != 0)
1523 {
1524 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1525 ptr++;
1526 #ifdef SUPPORT_UTF
1527 if (utf) FORWARDCHAR(ptr);
1528 #endif
1529 }
1530 if (*ptr == 0) goto FAIL_EXIT;
1531 continue;
1532 }
1533
1534 /* Check for the special metacharacters */
1535
1536 if (*ptr == CHAR_LEFT_PARENTHESIS)
1537 {
1538 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1539 if (rc > 0) return rc;
1540 if (*ptr == 0) goto FAIL_EXIT;
1541 }
1542
1543 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1544 {
1545 if (dup_parens && *count < hwm_count) *count = hwm_count;
1546 goto FAIL_EXIT;
1547 }
1548
1549 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1550 {
1551 if (*count > hwm_count) hwm_count = *count;
1552 *count = start_count;
1553 }
1554 }
1555
1556 FAIL_EXIT:
1557 *ptrptr = ptr;
1558 return -1;
1559 }
1560
1561
1562
1563
1564 /*************************************************
1565 * Find forward referenced subpattern *
1566 *************************************************/
1567
1568 /* This function scans along a pattern's text looking for capturing
1569 subpatterns, and counting them. If it finds a named pattern that matches the
1570 name it is given, it returns its number. Alternatively, if the name is NULL, it
1571 returns when it reaches a given numbered subpattern. This is used for forward
1572 references to subpatterns. We used to be able to start this scan from the
1573 current compiling point, using the current count value from cd->bracount, and
1574 do it all in a single loop, but the addition of the possibility of duplicate
1575 subpattern numbers means that we have to scan from the very start, in order to
1576 take account of such duplicates, and to use a recursive function to keep track
1577 of the different types of group.
1578
1579 Arguments:
1580 cd compile background data
1581 name name to seek, or NULL if seeking a numbered subpattern
1582 lorn name length, or subpattern number if name is NULL
1583 xmode TRUE if we are in /x mode
1584 utf TRUE if we are in UTF-8 / UTF-16 mode
1585
1586 Returns: the number of the found subpattern, or -1 if not found
1587 */
1588
1589 static int
1590 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1591 BOOL utf)
1592 {
1593 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1594 int count = 0;
1595 int rc;
1596
1597 /* If the pattern does not start with an opening parenthesis, the first call
1598 to find_parens_sub() will scan right to the end (if necessary). However, if it
1599 does start with a parenthesis, find_parens_sub() will return when it hits the
1600 matching closing parens. That is why we have to have a loop. */
1601
1602 for (;;)
1603 {
1604 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1605 if (rc > 0 || *ptr++ == 0) break;
1606 }
1607
1608 return rc;
1609 }
1610
1611
1612
1613
1614 /*************************************************
1615 * Find first significant op code *
1616 *************************************************/
1617
1618 /* This is called by several functions that scan a compiled expression looking
1619 for a fixed first character, or an anchoring op code etc. It skips over things
1620 that do not influence this. For some calls, it makes sense to skip negative
1621 forward and all backward assertions, and also the \b assertion; for others it
1622 does not.
1623
1624 Arguments:
1625 code pointer to the start of the group
1626 skipassert TRUE if certain assertions are to be skipped
1627
1628 Returns: pointer to the first significant opcode
1629 */
1630
1631 static const pcre_uchar*
1632 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1633 {
1634 for (;;)
1635 {
1636 switch ((int)*code)
1637 {
1638 case OP_ASSERT_NOT:
1639 case OP_ASSERTBACK:
1640 case OP_ASSERTBACK_NOT:
1641 if (!skipassert) return code;
1642 do code += GET(code, 1); while (*code == OP_ALT);
1643 code += PRIV(OP_lengths)[*code];
1644 break;
1645
1646 case OP_WORD_BOUNDARY:
1647 case OP_NOT_WORD_BOUNDARY:
1648 if (!skipassert) return code;
1649 /* Fall through */
1650
1651 case OP_CALLOUT:
1652 case OP_CREF:
1653 case OP_NCREF:
1654 case OP_RREF:
1655 case OP_NRREF:
1656 case OP_DEF:
1657 code += PRIV(OP_lengths)[*code];
1658 break;
1659
1660 default:
1661 return code;
1662 }
1663 }
1664 /* Control never reaches here */
1665 }
1666
1667
1668
1669
1670 /*************************************************
1671 * Find the fixed length of a branch *
1672 *************************************************/
1673
1674 /* Scan a branch and compute the fixed length of subject that will match it,
1675 if the length is fixed. This is needed for dealing with backward assertions.
1676 In UTF8 mode, the result is in characters rather than bytes. The branch is
1677 temporarily terminated with OP_END when this function is called.
1678
1679 This function is called when a backward assertion is encountered, so that if it
1680 fails, the error message can point to the correct place in the pattern.
1681 However, we cannot do this when the assertion contains subroutine calls,
1682 because they can be forward references. We solve this by remembering this case
1683 and doing the check at the end; a flag specifies which mode we are running in.
1684
1685 Arguments:
1686 code points to the start of the pattern (the bracket)
1687 utf TRUE in UTF-8 / UTF-16 mode
1688 atend TRUE if called when the pattern is complete
1689 cd the "compile data" structure
1690
1691 Returns: the fixed length,
1692 or -1 if there is no fixed length,
1693 or -2 if \C was encountered (in UTF-8 mode only)
1694 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1695 or -4 if an unknown opcode was encountered (internal error)
1696 */
1697
1698 static int
1699 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1700 {
1701 int length = -1;
1702
1703 register int branchlength = 0;
1704 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1705
1706 /* Scan along the opcodes for this branch. If we get to the end of the
1707 branch, check the length against that of the other branches. */
1708
1709 for (;;)
1710 {
1711 int d;
1712 pcre_uchar *ce, *cs;
1713 register int op = *cc;
1714
1715 switch (op)
1716 {
1717 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1718 OP_BRA (normal non-capturing bracket) because the other variants of these
1719 opcodes are all concerned with unlimited repeated groups, which of course
1720 are not of fixed length. */
1721
1722 case OP_CBRA:
1723 case OP_BRA:
1724 case OP_ONCE:
1725 case OP_ONCE_NC:
1726 case OP_COND:
1727 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1728 if (d < 0) return d;
1729 branchlength += d;
1730 do cc += GET(cc, 1); while (*cc == OP_ALT);
1731 cc += 1 + LINK_SIZE;
1732 break;
1733
1734 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1735 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1736 an ALT. If it is END it's the end of the outer call. All can be handled by
1737 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1738 because they all imply an unlimited repeat. */
1739
1740 case OP_ALT:
1741 case OP_KET:
1742 case OP_END:
1743 case OP_ACCEPT:
1744 case OP_ASSERT_ACCEPT:
1745 if (length < 0) length = branchlength;
1746 else if (length != branchlength) return -1;
1747 if (*cc != OP_ALT) return length;
1748 cc += 1 + LINK_SIZE;
1749 branchlength = 0;
1750 break;
1751
1752 /* A true recursion implies not fixed length, but a subroutine call may
1753 be OK. If the subroutine is a forward reference, we can't deal with
1754 it until the end of the pattern, so return -3. */
1755
1756 case OP_RECURSE:
1757 if (!atend) return -3;
1758 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1759 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1760 if (cc > cs && cc < ce) return -1; /* Recursion */
1761 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1762 if (d < 0) return d;
1763 branchlength += d;
1764 cc += 1 + LINK_SIZE;
1765 break;
1766
1767 /* Skip over assertive subpatterns */
1768
1769 case OP_ASSERT:
1770 case OP_ASSERT_NOT:
1771 case OP_ASSERTBACK:
1772 case OP_ASSERTBACK_NOT:
1773 do cc += GET(cc, 1); while (*cc == OP_ALT);
1774 cc += PRIV(OP_lengths)[*cc];
1775 break;
1776
1777 /* Skip over things that don't match chars */
1778
1779 case OP_MARK:
1780 case OP_PRUNE_ARG:
1781 case OP_SKIP_ARG:
1782 case OP_THEN_ARG:
1783 cc += cc[1] + PRIV(OP_lengths)[*cc];
1784 break;
1785
1786 case OP_CALLOUT:
1787 case OP_CIRC:
1788 case OP_CIRCM:
1789 case OP_CLOSE:
1790 case OP_COMMIT:
1791 case OP_CREF:
1792 case OP_DEF:
1793 case OP_DOLL:
1794 case OP_DOLLM:
1795 case OP_EOD:
1796 case OP_EODN:
1797 case OP_FAIL:
1798 case OP_NCREF:
1799 case OP_NRREF:
1800 case OP_NOT_WORD_BOUNDARY:
1801 case OP_PRUNE:
1802 case OP_REVERSE:
1803 case OP_RREF:
1804 case OP_SET_SOM:
1805 case OP_SKIP:
1806 case OP_SOD:
1807 case OP_SOM:
1808 case OP_THEN:
1809 case OP_WORD_BOUNDARY:
1810 cc += PRIV(OP_lengths)[*cc];
1811 break;
1812
1813 /* Handle literal characters */
1814
1815 case OP_CHAR:
1816 case OP_CHARI:
1817 case OP_NOT:
1818 case OP_NOTI:
1819 branchlength++;
1820 cc += 2;
1821 #ifdef SUPPORT_UTF
1822 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1823 #endif
1824 break;
1825
1826 /* Handle exact repetitions. The count is already in characters, but we
1827 need to skip over a multibyte character in UTF8 mode. */
1828
1829 case OP_EXACT:
1830 case OP_EXACTI:
1831 case OP_NOTEXACT:
1832 case OP_NOTEXACTI:
1833 branchlength += GET2(cc,1);
1834 cc += 2 + IMM2_SIZE;
1835 #ifdef SUPPORT_UTF
1836 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1837 #endif
1838 break;
1839
1840 case OP_TYPEEXACT:
1841 branchlength += GET2(cc,1);
1842 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1843 cc += 1 + IMM2_SIZE + 1;
1844 break;
1845
1846 /* Handle single-char matchers */
1847
1848 case OP_PROP:
1849 case OP_NOTPROP:
1850 cc += 2;
1851 /* Fall through */
1852
1853 case OP_HSPACE:
1854 case OP_VSPACE:
1855 case OP_NOT_HSPACE:
1856 case OP_NOT_VSPACE:
1857 case OP_NOT_DIGIT:
1858 case OP_DIGIT:
1859 case OP_NOT_WHITESPACE:
1860 case OP_WHITESPACE:
1861 case OP_NOT_WORDCHAR:
1862 case OP_WORDCHAR:
1863 case OP_ANY:
1864 case OP_ALLANY:
1865 branchlength++;
1866 cc++;
1867 break;
1868
1869 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1870 otherwise \C is coded as OP_ALLANY. */
1871
1872 case OP_ANYBYTE:
1873 return -2;
1874
1875 /* Check a class for variable quantification */
1876
1877 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1878 case OP_XCLASS:
1879 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1880 /* Fall through */
1881 #endif
1882
1883 case OP_CLASS:
1884 case OP_NCLASS:
1885 cc += PRIV(OP_lengths)[OP_CLASS];
1886
1887 switch (*cc)
1888 {
1889 case OP_CRPLUS:
1890 case OP_CRMINPLUS:
1891 case OP_CRSTAR:
1892 case OP_CRMINSTAR:
1893 case OP_CRQUERY:
1894 case OP_CRMINQUERY:
1895 return -1;
1896
1897 case OP_CRRANGE:
1898 case OP_CRMINRANGE:
1899 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1900 branchlength += GET2(cc,1);
1901 cc += 1 + 2 * IMM2_SIZE;
1902 break;
1903
1904 default:
1905 branchlength++;
1906 }
1907 break;
1908
1909 /* Anything else is variable length */
1910
1911 case OP_ANYNL:
1912 case OP_BRAMINZERO:
1913 case OP_BRAPOS:
1914 case OP_BRAPOSZERO:
1915 case OP_BRAZERO:
1916 case OP_CBRAPOS:
1917 case OP_EXTUNI:
1918 case OP_KETRMAX:
1919 case OP_KETRMIN:
1920 case OP_KETRPOS:
1921 case OP_MINPLUS:
1922 case OP_MINPLUSI:
1923 case OP_MINQUERY:
1924 case OP_MINQUERYI:
1925 case OP_MINSTAR:
1926 case OP_MINSTARI:
1927 case OP_MINUPTO:
1928 case OP_MINUPTOI:
1929 case OP_NOTMINPLUS:
1930 case OP_NOTMINPLUSI:
1931 case OP_NOTMINQUERY:
1932 case OP_NOTMINQUERYI:
1933 case OP_NOTMINSTAR:
1934 case OP_NOTMINSTARI:
1935 case OP_NOTMINUPTO:
1936 case OP_NOTMINUPTOI:
1937 case OP_NOTPLUS:
1938 case OP_NOTPLUSI:
1939 case OP_NOTPOSPLUS:
1940 case OP_NOTPOSPLUSI:
1941 case OP_NOTPOSQUERY:
1942 case OP_NOTPOSQUERYI:
1943 case OP_NOTPOSSTAR:
1944 case OP_NOTPOSSTARI:
1945 case OP_NOTPOSUPTO:
1946 case OP_NOTPOSUPTOI:
1947 case OP_NOTQUERY:
1948 case OP_NOTQUERYI:
1949 case OP_NOTSTAR:
1950 case OP_NOTSTARI:
1951 case OP_NOTUPTO:
1952 case OP_NOTUPTOI:
1953 case OP_PLUS:
1954 case OP_PLUSI:
1955 case OP_POSPLUS:
1956 case OP_POSPLUSI:
1957 case OP_POSQUERY:
1958 case OP_POSQUERYI:
1959 case OP_POSSTAR:
1960 case OP_POSSTARI:
1961 case OP_POSUPTO:
1962 case OP_POSUPTOI:
1963 case OP_QUERY:
1964 case OP_QUERYI:
1965 case OP_REF:
1966 case OP_REFI:
1967 case OP_SBRA:
1968 case OP_SBRAPOS:
1969 case OP_SCBRA:
1970 case OP_SCBRAPOS:
1971 case OP_SCOND:
1972 case OP_SKIPZERO:
1973 case OP_STAR:
1974 case OP_STARI:
1975 case OP_TYPEMINPLUS:
1976 case OP_TYPEMINQUERY:
1977 case OP_TYPEMINSTAR:
1978 case OP_TYPEMINUPTO:
1979 case OP_TYPEPLUS:
1980 case OP_TYPEPOSPLUS:
1981 case OP_TYPEPOSQUERY:
1982 case OP_TYPEPOSSTAR:
1983 case OP_TYPEPOSUPTO:
1984 case OP_TYPEQUERY:
1985 case OP_TYPESTAR:
1986 case OP_TYPEUPTO:
1987 case OP_UPTO:
1988 case OP_UPTOI:
1989 return -1;
1990
1991 /* Catch unrecognized opcodes so that when new ones are added they
1992 are not forgotten, as has happened in the past. */
1993
1994 default:
1995 return -4;
1996 }
1997 }
1998 /* Control never gets here */
1999 }
2000
2001
2002
2003
2004 /*************************************************
2005 * Scan compiled regex for specific bracket *
2006 *************************************************/
2007
2008 /* This little function scans through a compiled pattern until it finds a
2009 capturing bracket with the given number, or, if the number is negative, an
2010 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2011 so that it can be called from pcre_study() when finding the minimum matching
2012 length.
2013
2014 Arguments:
2015 code points to start of expression
2016 utf TRUE in UTF-8 / UTF-16 mode
2017 number the required bracket number or negative to find a lookbehind
2018
2019 Returns: pointer to the opcode for the bracket, or NULL if not found
2020 */
2021
2022 const pcre_uchar *
2023 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2024 {
2025 for (;;)
2026 {
2027 register int c = *code;
2028
2029 if (c == OP_END) return NULL;
2030
2031 /* XCLASS is used for classes that cannot be represented just by a bit
2032 map. This includes negated single high-valued characters. The length in
2033 the table is zero; the actual length is stored in the compiled code. */
2034
2035 if (c == OP_XCLASS) code += GET(code, 1);
2036
2037 /* Handle recursion */
2038
2039 else if (c == OP_REVERSE)
2040 {
2041 if (number < 0) return (pcre_uchar *)code;
2042 code += PRIV(OP_lengths)[c];
2043 }
2044
2045 /* Handle capturing bracket */
2046
2047 else if (c == OP_CBRA || c == OP_SCBRA ||
2048 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2049 {
2050 int n = GET2(code, 1+LINK_SIZE);
2051 if (n == number) return (pcre_uchar *)code;
2052 code += PRIV(OP_lengths)[c];
2053 }
2054
2055 /* Otherwise, we can get the item's length from the table, except that for
2056 repeated character types, we have to test for \p and \P, which have an extra
2057 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2058 must add in its length. */
2059
2060 else
2061 {
2062 switch(c)
2063 {
2064 case OP_TYPESTAR:
2065 case OP_TYPEMINSTAR:
2066 case OP_TYPEPLUS:
2067 case OP_TYPEMINPLUS:
2068 case OP_TYPEQUERY:
2069 case OP_TYPEMINQUERY:
2070 case OP_TYPEPOSSTAR:
2071 case OP_TYPEPOSPLUS:
2072 case OP_TYPEPOSQUERY:
2073 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2074 break;
2075
2076 case OP_TYPEUPTO:
2077 case OP_TYPEMINUPTO:
2078 case OP_TYPEEXACT:
2079 case OP_TYPEPOSUPTO:
2080 if (code[1 + IMM2_SIZE] == OP_PROP
2081 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2082 break;
2083
2084 case OP_MARK:
2085 case OP_PRUNE_ARG:
2086 case OP_SKIP_ARG:
2087 code += code[1];
2088 break;
2089
2090 case OP_THEN_ARG:
2091 code += code[1];
2092 break;
2093 }
2094
2095 /* Add in the fixed length from the table */
2096
2097 code += PRIV(OP_lengths)[c];
2098
2099 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2100 a multi-byte character. The length in the table is a minimum, so we have to
2101 arrange to skip the extra bytes. */
2102
2103 #ifdef SUPPORT_UTF
2104 if (utf) switch(c)
2105 {
2106 case OP_CHAR:
2107 case OP_CHARI:
2108 case OP_EXACT:
2109 case OP_EXACTI:
2110 case OP_UPTO:
2111 case OP_UPTOI:
2112 case OP_MINUPTO:
2113 case OP_MINUPTOI:
2114 case OP_POSUPTO:
2115 case OP_POSUPTOI:
2116 case OP_STAR:
2117 case OP_STARI:
2118 case OP_MINSTAR:
2119 case OP_MINSTARI:
2120 case OP_POSSTAR:
2121 case OP_POSSTARI:
2122 case OP_PLUS:
2123 case OP_PLUSI:
2124 case OP_MINPLUS:
2125 case OP_MINPLUSI:
2126 case OP_POSPLUS:
2127 case OP_POSPLUSI:
2128 case OP_QUERY:
2129 case OP_QUERYI:
2130 case OP_MINQUERY:
2131 case OP_MINQUERYI:
2132 case OP_POSQUERY:
2133 case OP_POSQUERYI:
2134 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2135 break;
2136 }
2137 #else
2138 (void)(utf); /* Keep compiler happy by referencing function argument */
2139 #endif
2140 }
2141 }
2142 }
2143
2144
2145
2146 /*************************************************
2147 * Scan compiled regex for recursion reference *
2148 *************************************************/
2149
2150 /* This little function scans through a compiled pattern until it finds an
2151 instance of OP_RECURSE.
2152
2153 Arguments:
2154 code points to start of expression
2155 utf TRUE in UTF-8 / UTF-16 mode
2156
2157 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2158 */
2159
2160 static const pcre_uchar *
2161 find_recurse(const pcre_uchar *code, BOOL utf)
2162 {
2163 for (;;)
2164 {
2165 register int c = *code;
2166 if (c == OP_END) return NULL;
2167 if (c == OP_RECURSE) return code;
2168
2169 /* XCLASS is used for classes that cannot be represented just by a bit
2170 map. This includes negated single high-valued characters. The length in
2171 the table is zero; the actual length is stored in the compiled code. */
2172
2173 if (c == OP_XCLASS) code += GET(code, 1);
2174
2175 /* Otherwise, we can get the item's length from the table, except that for
2176 repeated character types, we have to test for \p and \P, which have an extra
2177 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2178 must add in its length. */
2179
2180 else
2181 {
2182 switch(c)
2183 {
2184 case OP_TYPESTAR:
2185 case OP_TYPEMINSTAR:
2186 case OP_TYPEPLUS:
2187 case OP_TYPEMINPLUS:
2188 case OP_TYPEQUERY:
2189 case OP_TYPEMINQUERY:
2190 case OP_TYPEPOSSTAR:
2191 case OP_TYPEPOSPLUS:
2192 case OP_TYPEPOSQUERY:
2193 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2194 break;
2195
2196 case OP_TYPEPOSUPTO:
2197 case OP_TYPEUPTO:
2198 case OP_TYPEMINUPTO:
2199 case OP_TYPEEXACT:
2200 if (code[1 + IMM2_SIZE] == OP_PROP
2201 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2202 break;
2203
2204 case OP_MARK:
2205 case OP_PRUNE_ARG:
2206 case OP_SKIP_ARG:
2207 code += code[1];
2208 break;
2209
2210 case OP_THEN_ARG:
2211 code += code[1];
2212 break;
2213 }
2214
2215 /* Add in the fixed length from the table */
2216
2217 code += PRIV(OP_lengths)[c];
2218
2219 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2220 by a multi-byte character. The length in the table is a minimum, so we have
2221 to arrange to skip the extra bytes. */
2222
2223 #ifdef SUPPORT_UTF
2224 if (utf) switch(c)
2225 {
2226 case OP_CHAR:
2227 case OP_CHARI:
2228 case OP_NOT:
2229 case OP_NOTI:
2230 case OP_EXACT:
2231 case OP_EXACTI:
2232 case OP_NOTEXACT:
2233 case OP_NOTEXACTI:
2234 case OP_UPTO:
2235 case OP_UPTOI:
2236 case OP_NOTUPTO:
2237 case OP_NOTUPTOI:
2238 case OP_MINUPTO:
2239 case OP_MINUPTOI:
2240 case OP_NOTMINUPTO:
2241 case OP_NOTMINUPTOI:
2242 case OP_POSUPTO:
2243 case OP_POSUPTOI:
2244 case OP_NOTPOSUPTO:
2245 case OP_NOTPOSUPTOI:
2246 case OP_STAR:
2247 case OP_STARI:
2248 case OP_NOTSTAR:
2249 case OP_NOTSTARI:
2250 case OP_MINSTAR:
2251 case OP_MINSTARI:
2252 case OP_NOTMINSTAR:
2253 case OP_NOTMINSTARI:
2254 case OP_POSSTAR:
2255 case OP_POSSTARI:
2256 case OP_NOTPOSSTAR:
2257 case OP_NOTPOSSTARI:
2258 case OP_PLUS:
2259 case OP_PLUSI:
2260 case OP_NOTPLUS:
2261 case OP_NOTPLUSI:
2262 case OP_MINPLUS:
2263 case OP_MINPLUSI:
2264 case OP_NOTMINPLUS:
2265 case OP_NOTMINPLUSI:
2266 case OP_POSPLUS:
2267 case OP_POSPLUSI:
2268 case OP_NOTPOSPLUS:
2269 case OP_NOTPOSPLUSI:
2270 case OP_QUERY:
2271 case OP_QUERYI:
2272 case OP_NOTQUERY:
2273 case OP_NOTQUERYI:
2274 case OP_MINQUERY:
2275 case OP_MINQUERYI:
2276 case OP_NOTMINQUERY:
2277 case OP_NOTMINQUERYI:
2278 case OP_POSQUERY:
2279 case OP_POSQUERYI:
2280 case OP_NOTPOSQUERY:
2281 case OP_NOTPOSQUERYI:
2282 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2283 break;
2284 }
2285 #else
2286 (void)(utf); /* Keep compiler happy by referencing function argument */
2287 #endif
2288 }
2289 }
2290 }
2291
2292
2293
2294 /*************************************************
2295 * Scan compiled branch for non-emptiness *
2296 *************************************************/
2297
2298 /* This function scans through a branch of a compiled pattern to see whether it
2299 can match the empty string or not. It is called from could_be_empty()
2300 below and from compile_branch() when checking for an unlimited repeat of a
2301 group that can match nothing. Note that first_significant_code() skips over
2302 backward and negative forward assertions when its final argument is TRUE. If we
2303 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2304 bracket whose current branch will already have been scanned.
2305
2306 Arguments:
2307 code points to start of search
2308 endcode points to where to stop
2309 utf TRUE if in UTF-8 / UTF-16 mode
2310 cd contains pointers to tables etc.
2311
2312 Returns: TRUE if what is matched could be empty
2313 */
2314
2315 static BOOL
2316 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2317 BOOL utf, compile_data *cd)
2318 {
2319 register int c;
2320 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2321 code < endcode;
2322 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2323 {
2324 const pcre_uchar *ccode;
2325
2326 c = *code;
2327
2328 /* Skip over forward assertions; the other assertions are skipped by
2329 first_significant_code() with a TRUE final argument. */
2330
2331 if (c == OP_ASSERT)
2332 {
2333 do code += GET(code, 1); while (*code == OP_ALT);
2334 c = *code;
2335 continue;
2336 }
2337
2338 /* For a recursion/subroutine call, if its end has been reached, which
2339 implies a backward reference subroutine call, we can scan it. If it's a
2340 forward reference subroutine call, we can't. To detect forward reference
2341 we have to scan up the list that is kept in the workspace. This function is
2342 called only when doing the real compile, not during the pre-compile that
2343 measures the size of the compiled pattern. */
2344
2345 if (c == OP_RECURSE)
2346 {
2347 const pcre_uchar *scode;
2348 BOOL empty_branch;
2349
2350 /* Test for forward reference */
2351
2352 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2353 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2354
2355 /* Not a forward reference, test for completed backward reference */
2356
2357 empty_branch = FALSE;
2358 scode = cd->start_code + GET(code, 1);
2359 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2360
2361 /* Completed backwards reference */
2362
2363 do
2364 {
2365 if (could_be_empty_branch(scode, endcode, utf, cd))
2366 {
2367 empty_branch = TRUE;
2368 break;
2369 }
2370 scode += GET(scode, 1);
2371 }
2372 while (*scode == OP_ALT);
2373
2374 if (!empty_branch) return FALSE; /* All branches are non-empty */
2375 continue;
2376 }
2377
2378 /* Groups with zero repeats can of course be empty; skip them. */
2379
2380 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2381 c == OP_BRAPOSZERO)
2382 {
2383 code += PRIV(OP_lengths)[c];
2384 do code += GET(code, 1); while (*code == OP_ALT);
2385 c = *code;
2386 continue;
2387 }
2388
2389 /* A nested group that is already marked as "could be empty" can just be
2390 skipped. */
2391
2392 if (c == OP_SBRA || c == OP_SBRAPOS ||
2393 c == OP_SCBRA || c == OP_SCBRAPOS)
2394 {
2395 do code += GET(code, 1); while (*code == OP_ALT);
2396 c = *code;
2397 continue;
2398 }
2399
2400 /* For other groups, scan the branches. */
2401
2402 if (c == OP_BRA || c == OP_BRAPOS ||
2403 c == OP_CBRA || c == OP_CBRAPOS ||
2404 c == OP_ONCE || c == OP_ONCE_NC ||
2405 c == OP_COND)
2406 {
2407 BOOL empty_branch;
2408 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2409
2410 /* If a conditional group has only one branch, there is a second, implied,
2411 empty branch, so just skip over the conditional, because it could be empty.
2412 Otherwise, scan the individual branches of the group. */
2413
2414 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2415 code += GET(code, 1);
2416 else
2417 {
2418 empty_branch = FALSE;
2419 do
2420 {
2421 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2422 empty_branch = TRUE;
2423 code += GET(code, 1);
2424 }
2425 while (*code == OP_ALT);
2426 if (!empty_branch) return FALSE; /* All branches are non-empty */
2427 }
2428
2429 c = *code;
2430 continue;
2431 }
2432
2433 /* Handle the other opcodes */
2434
2435 switch (c)
2436 {
2437 /* Check for quantifiers after a class. XCLASS is used for classes that
2438 cannot be represented just by a bit map. This includes negated single
2439 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2440 actual length is stored in the compiled code, so we must update "code"
2441 here. */
2442
2443 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2444 case OP_XCLASS:
2445 ccode = code += GET(code, 1);
2446 goto CHECK_CLASS_REPEAT;
2447 #endif
2448
2449 case OP_CLASS:
2450 case OP_NCLASS:
2451 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2452
2453 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2454 CHECK_CLASS_REPEAT:
2455 #endif
2456
2457 switch (*ccode)
2458 {
2459 case OP_CRSTAR: /* These could be empty; continue */
2460 case OP_CRMINSTAR:
2461 case OP_CRQUERY:
2462 case OP_CRMINQUERY:
2463 break;
2464
2465 default: /* Non-repeat => class must match */
2466 case OP_CRPLUS: /* These repeats aren't empty */
2467 case OP_CRMINPLUS:
2468 return FALSE;
2469
2470 case OP_CRRANGE:
2471 case OP_CRMINRANGE:
2472 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2473 break;
2474 }
2475 break;
2476
2477 /* Opcodes that must match a character */
2478
2479 case OP_PROP:
2480 case OP_NOTPROP:
2481 case OP_EXTUNI:
2482 case OP_NOT_DIGIT:
2483 case OP_DIGIT:
2484 case OP_NOT_WHITESPACE:
2485 case OP_WHITESPACE:
2486 case OP_NOT_WORDCHAR:
2487 case OP_WORDCHAR:
2488 case OP_ANY:
2489 case OP_ALLANY:
2490 case OP_ANYBYTE:
2491 case OP_CHAR:
2492 case OP_CHARI:
2493 case OP_NOT:
2494 case OP_NOTI:
2495 case OP_PLUS:
2496 case OP_MINPLUS:
2497 case OP_POSPLUS:
2498 case OP_EXACT:
2499 case OP_NOTPLUS:
2500 case OP_NOTMINPLUS:
2501 case OP_NOTPOSPLUS:
2502 case OP_NOTEXACT:
2503 case OP_TYPEPLUS:
2504 case OP_TYPEMINPLUS:
2505 case OP_TYPEPOSPLUS:
2506 case OP_TYPEEXACT:
2507 return FALSE;
2508
2509 /* These are going to continue, as they may be empty, but we have to
2510 fudge the length for the \p and \P cases. */
2511
2512 case OP_TYPESTAR:
2513 case OP_TYPEMINSTAR:
2514 case OP_TYPEPOSSTAR:
2515 case OP_TYPEQUERY:
2516 case OP_TYPEMINQUERY:
2517 case OP_TYPEPOSQUERY:
2518 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2519 break;
2520
2521 /* Same for these */
2522
2523 case OP_TYPEUPTO:
2524 case OP_TYPEMINUPTO:
2525 case OP_TYPEPOSUPTO:
2526 if (code[1 + IMM2_SIZE] == OP_PROP
2527 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2528 break;
2529
2530 /* End of branch */
2531
2532 case OP_KET:
2533 case OP_KETRMAX:
2534 case OP_KETRMIN:
2535 case OP_KETRPOS:
2536 case OP_ALT:
2537 return TRUE;
2538
2539 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2540 MINUPTO, and POSUPTO may be followed by a multibyte character */
2541
2542 #ifdef SUPPORT_UTF
2543 case OP_STAR:
2544 case OP_STARI:
2545 case OP_MINSTAR:
2546 case OP_MINSTARI:
2547 case OP_POSSTAR:
2548 case OP_POSSTARI:
2549 case OP_QUERY:
2550 case OP_QUERYI:
2551 case OP_MINQUERY:
2552 case OP_MINQUERYI:
2553 case OP_POSQUERY:
2554 case OP_POSQUERYI:
2555 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2556 break;
2557
2558 case OP_UPTO:
2559 case OP_UPTOI:
2560 case OP_MINUPTO:
2561 case OP_MINUPTOI:
2562 case OP_POSUPTO:
2563 case OP_POSUPTOI:
2564 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2565 break;
2566 #endif
2567
2568 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2569 string. */
2570
2571 case OP_MARK:
2572 case OP_PRUNE_ARG:
2573 case OP_SKIP_ARG:
2574 code += code[1];
2575 break;
2576
2577 case OP_THEN_ARG:
2578 code += code[1];
2579 break;
2580
2581 /* None of the remaining opcodes are required to match a character. */
2582
2583 default:
2584 break;
2585 }
2586 }
2587
2588 return TRUE;
2589 }
2590
2591
2592
2593 /*************************************************
2594 * Scan compiled regex for non-emptiness *
2595 *************************************************/
2596
2597 /* This function is called to check for left recursive calls. We want to check
2598 the current branch of the current pattern to see if it could match the empty
2599 string. If it could, we must look outwards for branches at other levels,
2600 stopping when we pass beyond the bracket which is the subject of the recursion.
2601 This function is called only during the real compile, not during the
2602 pre-compile.
2603
2604 Arguments:
2605 code points to start of the recursion
2606 endcode points to where to stop (current RECURSE item)
2607 bcptr points to the chain of current (unclosed) branch starts
2608 utf TRUE if in UTF-8 / UTF-16 mode
2609 cd pointers to tables etc
2610
2611 Returns: TRUE if what is matched could be empty
2612 */
2613
2614 static BOOL
2615 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2616 branch_chain *bcptr, BOOL utf, compile_data *cd)
2617 {
2618 while (bcptr != NULL && bcptr->current_branch >= code)
2619 {
2620 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2621 return FALSE;
2622 bcptr = bcptr->outer;
2623 }
2624 return TRUE;
2625 }
2626
2627
2628
2629 /*************************************************
2630 * Check for POSIX class syntax *
2631 *************************************************/
2632
2633 /* This function is called when the sequence "[:" or "[." or "[=" is
2634 encountered in a character class. It checks whether this is followed by a
2635 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2636 reach an unescaped ']' without the special preceding character, return FALSE.
2637
2638 Originally, this function only recognized a sequence of letters between the
2639 terminators, but it seems that Perl recognizes any sequence of characters,
2640 though of course unknown POSIX names are subsequently rejected. Perl gives an
2641 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2642 didn't consider this to be a POSIX class. Likewise for [:1234:].
2643
2644 The problem in trying to be exactly like Perl is in the handling of escapes. We
2645 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2646 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2647 below handles the special case of \], but does not try to do any other escape
2648 processing. This makes it different from Perl for cases such as [:l\ower:]
2649 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2650 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2651 I think.
2652
2653 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2654 It seems that the appearance of a nested POSIX class supersedes an apparent
2655 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2656 a digit.
2657
2658 In Perl, unescaped square brackets may also appear as part of class names. For
2659 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2660 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2661 seem right at all. PCRE does not allow closing square brackets in POSIX class
2662 names.
2663
2664 Arguments:
2665 ptr pointer to the initial [
2666 endptr where to return the end pointer
2667
2668 Returns: TRUE or FALSE
2669 */
2670
2671 static BOOL
2672 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2673 {
2674 int terminator; /* Don't combine these lines; the Solaris cc */
2675 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2676 for (++ptr; *ptr != 0; ptr++)
2677 {
2678 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2679 ptr++;
2680 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2681 else
2682 {
2683 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2684 {
2685 *endptr = ptr;
2686 return TRUE;
2687 }
2688 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2689 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2690 ptr[1] == CHAR_EQUALS_SIGN) &&
2691 check_posix_syntax(ptr, endptr))
2692 return FALSE;
2693 }
2694 }
2695 return FALSE;
2696 }
2697
2698
2699
2700
2701 /*************************************************
2702 * Check POSIX class name *
2703 *************************************************/
2704
2705 /* This function is called to check the name given in a POSIX-style class entry
2706 such as [:alnum:].
2707
2708 Arguments:
2709 ptr points to the first letter
2710 len the length of the name
2711
2712 Returns: a value representing the name, or -1 if unknown
2713 */
2714
2715 static int
2716 check_posix_name(const pcre_uchar *ptr, int len)
2717 {
2718 const char *pn = posix_names;
2719 register int yield = 0;
2720 while (posix_name_lengths[yield] != 0)
2721 {
2722 if (len == posix_name_lengths[yield] &&
2723 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2724 pn += posix_name_lengths[yield] + 1;
2725 yield++;
2726 }
2727 return -1;
2728 }
2729
2730
2731 /*************************************************
2732 * Adjust OP_RECURSE items in repeated group *
2733 *************************************************/
2734
2735 /* OP_RECURSE items contain an offset from the start of the regex to the group
2736 that is referenced. This means that groups can be replicated for fixed
2737 repetition simply by copying (because the recursion is allowed to refer to
2738 earlier groups that are outside the current group). However, when a group is
2739 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2740 inserted before it, after it has been compiled. This means that any OP_RECURSE
2741 items within it that refer to the group itself or any contained groups have to
2742 have their offsets adjusted. That one of the jobs of this function. Before it
2743 is called, the partially compiled regex must be temporarily terminated with
2744 OP_END.
2745
2746 This function has been extended with the possibility of forward references for
2747 recursions and subroutine calls. It must also check the list of such references
2748 for the group we are dealing with. If it finds that one of the recursions in
2749 the current group is on this list, it adjusts the offset in the list, not the
2750 value in the reference (which is a group number).
2751
2752 Arguments:
2753 group points to the start of the group
2754 adjust the amount by which the group is to be moved
2755 utf TRUE in UTF-8 / UTF-16 mode
2756 cd contains pointers to tables etc.
2757 save_hwm the hwm forward reference pointer at the start of the group
2758
2759 Returns: nothing
2760 */
2761
2762 static void
2763 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2764 pcre_uchar *save_hwm)
2765 {
2766 pcre_uchar *ptr = group;
2767
2768 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2769 {
2770 int offset;
2771 pcre_uchar *hc;
2772
2773 /* See if this recursion is on the forward reference list. If so, adjust the
2774 reference. */
2775
2776 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2777 {
2778 offset = GET(hc, 0);
2779 if (cd->start_code + offset == ptr + 1)
2780 {
2781 PUT(hc, 0, offset + adjust);
2782 break;
2783 }
2784 }
2785
2786 /* Otherwise, adjust the recursion offset if it's after the start of this
2787 group. */
2788
2789 if (hc >= cd->hwm)
2790 {
2791 offset = GET(ptr, 1);
2792 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2793 }
2794
2795 ptr += 1 + LINK_SIZE;
2796 }
2797 }
2798
2799
2800
2801 /*************************************************
2802 * Insert an automatic callout point *
2803 *************************************************/
2804
2805 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2806 callout points before each pattern item.
2807
2808 Arguments:
2809 code current code pointer
2810 ptr current pattern pointer
2811 cd pointers to tables etc
2812
2813 Returns: new code pointer
2814 */
2815
2816 static pcre_uchar *
2817 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2818 {
2819 *code++ = OP_CALLOUT;
2820 *code++ = 255;
2821 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2822 PUT(code, LINK_SIZE, 0); /* Default length */
2823 return code + 2 * LINK_SIZE;
2824 }
2825
2826
2827
2828 /*************************************************
2829 * Complete a callout item *
2830 *************************************************/
2831
2832 /* A callout item contains the length of the next item in the pattern, which
2833 we can't fill in till after we have reached the relevant point. This is used
2834 for both automatic and manual callouts.
2835
2836 Arguments:
2837 previous_callout points to previous callout item
2838 ptr current pattern pointer
2839 cd pointers to tables etc
2840
2841 Returns: nothing
2842 */
2843
2844 static void
2845 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2846 {
2847 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2848 PUT(previous_callout, 2 + LINK_SIZE, length);
2849 }
2850
2851
2852
2853 #ifdef SUPPORT_UCP
2854 /*************************************************
2855 * Get othercase range *
2856 *************************************************/
2857
2858 /* This function is passed the start and end of a class range, in UTF-8 mode
2859 with UCP support. It searches up the characters, looking for internal ranges of
2860 characters in the "other" case. Each call returns the next one, updating the
2861 start address.
2862
2863 Arguments:
2864 cptr points to starting character value; updated
2865 d end value
2866 ocptr where to put start of othercase range
2867 odptr where to put end of othercase range
2868
2869 Yield: TRUE when range returned; FALSE when no more
2870 */
2871
2872 static BOOL
2873 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2874 unsigned int *odptr)
2875 {
2876 unsigned int c, othercase, next;
2877
2878 for (c = *cptr; c <= d; c++)
2879 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2880
2881 if (c > d) return FALSE;
2882
2883 *ocptr = othercase;
2884 next = othercase + 1;
2885
2886 for (++c; c <= d; c++)
2887 {
2888 if (UCD_OTHERCASE(c) != next) break;
2889 next++;
2890 }
2891
2892 *odptr = next - 1;
2893 *cptr = c;
2894
2895 return TRUE;
2896 }
2897
2898
2899
2900 /*************************************************
2901 * Check a character and a property *
2902 *************************************************/
2903
2904 /* This function is called by check_auto_possessive() when a property item
2905 is adjacent to a fixed character.
2906
2907 Arguments:
2908 c the character
2909 ptype the property type
2910 pdata the data for the type
2911 negated TRUE if it's a negated property (\P or \p{^)
2912
2913 Returns: TRUE if auto-possessifying is OK
2914 */
2915
2916 static BOOL
2917 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2918 {
2919 const ucd_record *prop = GET_UCD(c);
2920 switch(ptype)
2921 {
2922 case PT_LAMP:
2923 return (prop->chartype == ucp_Lu ||
2924 prop->chartype == ucp_Ll ||
2925 prop->chartype == ucp_Lt) == negated;
2926
2927 case PT_GC:
2928 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2929
2930 case PT_PC:
2931 return (pdata == prop->chartype) == negated;
2932
2933 case PT_SC:
2934 return (pdata == prop->script) == negated;
2935
2936 /* These are specials */
2937
2938 case PT_ALNUM:
2939 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2940 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2941
2942 case PT_SPACE: /* Perl space */
2943 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2944 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2945 == negated;
2946
2947 case PT_PXSPACE: /* POSIX space */
2948 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2949 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2950 c == CHAR_FF || c == CHAR_CR)
2951 == negated;
2952
2953 case PT_WORD:
2954 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2955 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2956 c == CHAR_UNDERSCORE) == negated;
2957 }
2958 return FALSE;
2959 }
2960 #endif /* SUPPORT_UCP */
2961
2962
2963
2964 /*************************************************
2965 * Check if auto-possessifying is possible *
2966 *************************************************/
2967
2968 /* This function is called for unlimited repeats of certain items, to see
2969 whether the next thing could possibly match the repeated item. If not, it makes
2970 sense to automatically possessify the repeated item.
2971
2972 Arguments:
2973 previous pointer to the repeated opcode
2974 utf TRUE in UTF-8 / UTF-16 mode
2975 ptr next character in pattern
2976 options options bits
2977 cd contains pointers to tables etc.
2978
2979 Returns: TRUE if possessifying is wanted
2980 */
2981
2982 static BOOL
2983 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2984 const pcre_uchar *ptr, int options, compile_data *cd)
2985 {
2986 pcre_int32 c, next;
2987 int op_code = *previous++;
2988
2989 /* Skip whitespace and comments in extended mode */
2990
2991 if ((options & PCRE_EXTENDED) != 0)
2992 {
2993 for (;;)
2994 {
2995 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2996 if (*ptr == CHAR_NUMBER_SIGN)
2997 {
2998 ptr++;
2999 while (*ptr != 0)
3000 {
3001 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3002 ptr++;
3003 #ifdef SUPPORT_UTF
3004 if (utf) FORWARDCHAR(ptr);
3005 #endif
3006 }
3007 }
3008 else break;
3009 }
3010 }
3011
3012 /* If the next item is one that we can handle, get its value. A non-negative
3013 value is a character, a negative value is an escape value. */
3014
3015 if (*ptr == CHAR_BACKSLASH)
3016 {
3017 int temperrorcode = 0;
3018 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
3019 if (temperrorcode != 0) return FALSE;
3020 ptr++; /* Point after the escape sequence */
3021 }
3022 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3023 {
3024 #ifdef SUPPORT_UTF
3025 if (utf) { GETCHARINC(next, ptr); } else
3026 #endif
3027 next = *ptr++;
3028 }
3029 else return FALSE;
3030
3031 /* Skip whitespace and comments in extended mode */
3032
3033 if ((options & PCRE_EXTENDED) != 0)
3034 {
3035 for (;;)
3036 {
3037 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3038 if (*ptr == CHAR_NUMBER_SIGN)
3039 {
3040 ptr++;
3041 while (*ptr != 0)
3042 {
3043 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3044 ptr++;
3045 #ifdef SUPPORT_UTF
3046 if (utf) FORWARDCHAR(ptr);
3047 #endif
3048 }
3049 }
3050 else break;
3051 }
3052 }
3053
3054 /* If the next thing is itself optional, we have to give up. */
3055
3056 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3057 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3058 return FALSE;
3059
3060 /* Now compare the next item with the previous opcode. First, handle cases when
3061 the next item is a character. */
3062
3063 if (next >= 0) switch(op_code)
3064 {
3065 case OP_CHAR:
3066 #ifdef SUPPORT_UTF
3067 GETCHARTEST(c, previous);
3068 #else
3069 c = *previous;
3070 #endif
3071 return c != next;
3072
3073 /* For CHARI (caseless character) we must check the other case. If we have
3074 Unicode property support, we can use it to test the other case of
3075 high-valued characters. */
3076
3077 case OP_CHARI:
3078 #ifdef SUPPORT_UTF
3079 GETCHARTEST(c, previous);
3080 #else
3081 c = *previous;
3082 #endif
3083 if (c == next) return FALSE;
3084 #ifdef SUPPORT_UTF
3085 if (utf)
3086 {
3087 unsigned int othercase;
3088 if (next < 128) othercase = cd->fcc[next]; else
3089 #ifdef SUPPORT_UCP
3090 othercase = UCD_OTHERCASE((unsigned int)next);
3091 #else
3092 othercase = NOTACHAR;
3093 #endif
3094 return (unsigned int)c != othercase;
3095 }
3096 else
3097 #endif /* SUPPORT_UTF */
3098 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3099
3100 case OP_NOT:
3101 #ifdef SUPPORT_UTF
3102 GETCHARTEST(c, previous);
3103 #else
3104 c = *previous;
3105 #endif
3106 return c == next;
3107
3108 case OP_NOTI:
3109 #ifdef SUPPORT_UTF
3110 GETCHARTEST(c, previous);
3111 #else
3112 c = *previous;
3113 #endif
3114 if (c == next) return TRUE;
3115 #ifdef SUPPORT_UTF
3116 if (utf)
3117 {
3118 unsigned int othercase;
3119 if (next < 128) othercase = cd->fcc[next]; else
3120 #ifdef SUPPORT_UCP
3121 othercase = UCD_OTHERCASE((unsigned int)next);
3122 #else
3123 othercase = NOTACHAR;
3124 #endif
3125 return (unsigned int)c == othercase;
3126 }
3127 else
3128 #endif /* SUPPORT_UTF */
3129 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3130
3131 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3132 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3133
3134 case OP_DIGIT:
3135 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3136
3137 case OP_NOT_DIGIT:
3138 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3139
3140 case OP_WHITESPACE:
3141 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3142
3143 case OP_NOT_WHITESPACE:
3144 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3145
3146 case OP_WORDCHAR:
3147 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3148
3149 case OP_NOT_WORDCHAR:
3150 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3151
3152 case OP_HSPACE:
3153 case OP_NOT_HSPACE:
3154 switch(next)
3155 {
3156 case 0x09:
3157 case 0x20:
3158 case 0xa0:
3159 case 0x1680:
3160 case 0x180e:
3161 case 0x2000:
3162 case 0x2001:
3163 case 0x2002:
3164 case 0x2003:
3165 case 0x2004:
3166 case 0x2005:
3167 case 0x2006:
3168 case 0x2007:
3169 case 0x2008:
3170 case 0x2009:
3171 case 0x200A:
3172 case 0x202f:
3173 case 0x205f:
3174 case 0x3000:
3175 return op_code == OP_NOT_HSPACE;
3176 default:
3177 return op_code != OP_NOT_HSPACE;
3178 }
3179
3180 case OP_ANYNL:
3181 case OP_VSPACE:
3182 case OP_NOT_VSPACE:
3183 switch(next)
3184 {
3185 case 0x0a:
3186 case 0x0b:
3187 case 0x0c:
3188 case 0x0d:
3189 case 0x85:
3190 case 0x2028:
3191 case 0x2029:
3192 return op_code == OP_NOT_VSPACE;
3193 default:
3194 return op_code != OP_NOT_VSPACE;
3195 }
3196
3197 #ifdef SUPPORT_UCP
3198 case OP_PROP:
3199 return check_char_prop(next, previous[0], previous[1], FALSE);
3200
3201 case OP_NOTPROP:
3202 return check_char_prop(next, previous[0], previous[1], TRUE);
3203 #endif
3204
3205 default:
3206 return FALSE;
3207 }
3208
3209
3210 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3211 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3212 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3213 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3214 replaced by OP_PROP codes when PCRE_UCP is set. */
3215
3216 switch(op_code)
3217 {
3218 case OP_CHAR:
3219 case OP_CHARI:
3220 #ifdef SUPPORT_UTF
3221 GETCHARTEST(c, previous);
3222 #else
3223 c = *previous;
3224 #endif
3225 switch(-next)
3226 {
3227 case ESC_d:
3228 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3229
3230 case ESC_D:
3231 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3232
3233 case ESC_s:
3234 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3235
3236 case ESC_S:
3237 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3238
3239 case ESC_w:
3240 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3241
3242 case ESC_W:
3243 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3244
3245 case ESC_h:
3246 case ESC_H:
3247 switch(c)
3248 {
3249 case 0x09:
3250 case 0x20:
3251 case 0xa0:
3252 case 0x1680:
3253 case 0x180e:
3254 case 0x2000:
3255 case 0x2001:
3256 case 0x2002:
3257 case 0x2003:
3258 case 0x2004:
3259 case 0x2005:
3260 case 0x2006:
3261 case 0x2007:
3262 case 0x2008:
3263 case 0x2009:
3264 case 0x200A:
3265 case 0x202f:
3266 case 0x205f:
3267 case 0x3000:
3268 return -next != ESC_h;
3269 default:
3270 return -next == ESC_h;
3271 }
3272
3273 case ESC_v:
3274 case ESC_V:
3275 switch(c)
3276 {
3277 case 0x0a:
3278 case 0x0b:
3279 case 0x0c:
3280 case 0x0d:
3281 case 0x85:
3282 case 0x2028:
3283 case 0x2029:
3284 return -next != ESC_v;
3285 default:
3286 return -next == ESC_v;
3287 }
3288
3289 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3290 their substitutions and process them. The result will always be either
3291 -ESC_p or -ESC_P. Then fall through to process those values. */
3292
3293 #ifdef SUPPORT_UCP
3294 case ESC_du:
3295 case ESC_DU:
3296 case ESC_wu:
3297 case ESC_WU:
3298 case ESC_su:
3299 case ESC_SU:
3300 {
3301 int temperrorcode = 0;
3302 ptr = substitutes[-next - ESC_DU];
3303 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3304 if (temperrorcode != 0) return FALSE;
3305 ptr++; /* For compatibility */
3306 }
3307 /* Fall through */
3308
3309 case ESC_p:
3310 case ESC_P:
3311 {
3312 int ptype, pdata, errorcodeptr;
3313 BOOL negated;
3314
3315 ptr--; /* Make ptr point at the p or P */
3316 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3317 if (ptype < 0) return FALSE;
3318 ptr++; /* Point past the final curly ket */
3319
3320 /* If the property item is optional, we have to give up. (When generated
3321 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3322 to the original \d etc. At this point, ptr will point to a zero byte. */
3323
3324 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3325 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3326 return FALSE;
3327
3328 /* Do the property check. */
3329
3330 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3331 }
3332 #endif
3333
3334 default:
3335 return FALSE;
3336 }
3337
3338 /* In principle, support for Unicode properties should be integrated here as
3339 well. It means re-organizing the above code so as to get hold of the property
3340 values before switching on the op-code. However, I wonder how many patterns
3341 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3342 these op-codes are never generated.) */
3343
3344 case OP_DIGIT:
3345 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3346 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3347
3348 case OP_NOT_DIGIT:
3349 return next == -ESC_d;
3350
3351 case OP_WHITESPACE:
3352 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3353
3354 case OP_NOT_WHITESPACE:
3355 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3356
3357 case OP_HSPACE:
3358 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3359 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3360
3361 case OP_NOT_HSPACE:
3362 return next == -ESC_h;
3363
3364 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3365 case OP_ANYNL:
3366 case OP_VSPACE:
3367 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3368
3369 case OP_NOT_VSPACE:
3370 return next == -ESC_v || next == -ESC_R;
3371
3372 case OP_WORDCHAR:
3373 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3374 next == -ESC_v || next == -ESC_R;
3375
3376 case OP_NOT_WORDCHAR:
3377 return next == -ESC_w || next == -ESC_d;
3378
3379 default:
3380 return FALSE;
3381 }
3382
3383 /* Control does not reach here */
3384 }
3385
3386
3387
3388 /*************************************************
3389 * Compile one branch *
3390 *************************************************/
3391
3392 /* Scan the pattern, compiling it into the a vector. If the options are
3393 changed during the branch, the pointer is used to change the external options
3394 bits. This function is used during the pre-compile phase when we are trying
3395 to find out the amount of memory needed, as well as during the real compile
3396 phase. The value of lengthptr distinguishes the two phases.
3397
3398 Arguments:
3399 optionsptr pointer to the option bits
3400 codeptr points to the pointer to the current code point
3401 ptrptr points to the current pattern pointer
3402 errorcodeptr points to error code variable
3403 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3404 reqcharptr set to the last literal character required, else < 0
3405 bcptr points to current branch chain
3406 cond_depth conditional nesting depth
3407 cd contains pointers to tables etc.
3408 lengthptr NULL during the real compile phase
3409 points to length accumulator during pre-compile phase
3410
3411 Returns: TRUE on success
3412 FALSE, with *errorcodeptr set non-zero on error
3413 */
3414
3415 static BOOL
3416 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3417 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3418 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3419 compile_data *cd, int *lengthptr)
3420 {
3421 int repeat_type, op_type;
3422 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3423 int bravalue = 0;
3424 int greedy_default, greedy_non_default;
3425 pcre_int32 firstchar, reqchar;
3426 pcre_int32 zeroreqchar, zerofirstchar;
3427 pcre_int32 req_caseopt, reqvary, tempreqvary;
3428 int options = *optionsptr; /* May change dynamically */
3429 int after_manual_callout = 0;
3430 int length_prevgroup = 0;
3431 register int c;
3432 register pcre_uchar *code = *codeptr;
3433 pcre_uchar *last_code = code;
3434 pcre_uchar *orig_code = code;
3435 pcre_uchar *tempcode;
3436 BOOL inescq = FALSE;
3437 BOOL groupsetfirstchar = FALSE;
3438 const pcre_uchar *ptr = *ptrptr;
3439 const pcre_uchar *tempptr;
3440 const pcre_uchar *nestptr = NULL;
3441 pcre_uchar *previous = NULL;
3442 pcre_uchar *previous_callout = NULL;
3443 pcre_uchar *save_hwm = NULL;
3444 pcre_uint8 classbits[32];
3445
3446 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3447 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3448 dynamically as we process the pattern. */
3449
3450 #ifdef SUPPORT_UTF
3451 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3452 BOOL utf = (options & PCRE_UTF8) != 0;
3453 pcre_uchar utf_chars[6];
3454 #else
3455 BOOL utf = FALSE;
3456 #endif
3457
3458 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3459
3460 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3461 BOOL xclass;
3462 pcre_uchar *class_uchardata;
3463 pcre_uchar *class_uchardata_base;
3464 #endif
3465
3466 #ifdef PCRE_DEBUG
3467 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3468 #endif
3469
3470 /* Set up the default and non-default settings for greediness */
3471
3472 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3473 greedy_non_default = greedy_default ^ 1;
3474
3475 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3476 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3477 matches a non-fixed char first char; reqchar just remains unset if we never
3478 find one.
3479
3480 When we hit a repeat whose minimum is zero, we may have to adjust these values
3481 to take the zero repeat into account. This is implemented by setting them to
3482 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3483 item types that can be repeated set these backoff variables appropriately. */
3484
3485 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3486
3487 /* The variable req_caseopt contains either the REQ_CASELESS value
3488 or zero, according to the current setting of the caseless flag. The
3489 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3490 firstchar or reqchar variables to record the case status of the
3491 value. This is used only for ASCII characters. */
3492
3493 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3494
3495 /* Switch on next character until the end of the branch */
3496
3497 for (;; ptr++)
3498 {
3499 BOOL negate_class;
3500 BOOL should_flip_negation;
3501 BOOL possessive_quantifier;
3502 BOOL is_quantifier;
3503 BOOL is_recurse;
3504 BOOL reset_bracount;
3505 int class_has_8bitchar;
3506 int class_single_char;
3507 int newoptions;
3508 int recno;
3509 int refsign;
3510 int skipbytes;
3511 int subreqchar;
3512 int subfirstchar;
3513 int terminator;
3514 int mclength;
3515 int tempbracount;
3516 pcre_uchar mcbuffer[8];
3517
3518 /* Get next character in the pattern */
3519
3520 c = *ptr;
3521
3522 /* If we are at the end of a nested substitution, revert to the outer level
3523 string. Nesting only happens one level deep. */
3524
3525 if (c == 0 && nestptr != NULL)
3526 {
3527 ptr = nestptr;
3528 nestptr = NULL;
3529 c = *ptr;
3530 }
3531
3532 /* If we are in the pre-compile phase, accumulate the length used for the
3533 previous cycle of this loop. */
3534
3535 if (lengthptr != NULL)
3536 {
3537 #ifdef PCRE_DEBUG
3538 if (code > cd->hwm) cd->hwm = code; /* High water info */
3539 #endif
3540 if (code > cd->start_workspace + cd->workspace_size -
3541 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3542 {
3543 *errorcodeptr = ERR52;
3544 goto FAILED;
3545 }
3546
3547 /* There is at least one situation where code goes backwards: this is the
3548 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3549 the class is simply eliminated. However, it is created first, so we have to
3550 allow memory for it. Therefore, don't ever reduce the length at this point.
3551 */
3552
3553 if (code < last_code) code = last_code;
3554
3555 /* Paranoid check for integer overflow */
3556
3557 if (OFLOW_MAX - *lengthptr < code - last_code)
3558 {
3559 *errorcodeptr = ERR20;
3560 goto FAILED;
3561 }
3562
3563 *lengthptr += (int)(code - last_code);
3564 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3565 (int)(code - last_code), c, c));
3566
3567 /* If "previous" is set and it is not at the start of the work space, move
3568 it back to there, in order to avoid filling up the work space. Otherwise,
3569 if "previous" is NULL, reset the current code pointer to the start. */
3570
3571 if (previous != NULL)
3572 {
3573 if (previous > orig_code)
3574 {
3575 memmove(orig_code, previous, IN_UCHARS(code - previous));
3576 code -= previous - orig_code;
3577 previous = orig_code;
3578 }
3579 }
3580 else code = orig_code;
3581
3582 /* Remember where this code item starts so we can pick up the length
3583 next time round. */
3584
3585 last_code = code;
3586 }
3587
3588 /* In the real compile phase, just check the workspace used by the forward
3589 reference list. */
3590
3591 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3592 WORK_SIZE_SAFETY_MARGIN)
3593 {
3594 *errorcodeptr = ERR52;
3595 goto FAILED;
3596 }
3597
3598 /* If in \Q...\E, check for the end; if not, we have a literal */
3599
3600 if (inescq && c != 0)
3601 {
3602 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3603 {
3604 inescq = FALSE;
3605 ptr++;
3606 continue;
3607 }
3608 else
3609 {
3610 if (previous_callout != NULL)
3611 {
3612 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3613 complete_callout(previous_callout, ptr, cd);
3614 previous_callout = NULL;
3615 }
3616 if ((options & PCRE_AUTO_CALLOUT) != 0)
3617 {
3618 previous_callout = code;
3619 code = auto_callout(code, ptr, cd);
3620 }
3621 goto NORMAL_CHAR;
3622 }
3623 }
3624
3625 /* Fill in length of a previous callout, except when the next thing is
3626 a quantifier. */
3627
3628 is_quantifier =
3629 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3630 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3631
3632 if (!is_quantifier && previous_callout != NULL &&
3633 after_manual_callout-- <= 0)
3634 {
3635 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3636 complete_callout(previous_callout, ptr, cd);
3637 previous_callout = NULL;
3638 }
3639
3640 /* In extended mode, skip white space and comments. */
3641
3642 if ((options & PCRE_EXTENDED) != 0)
3643 {
3644 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3645 if (c == CHAR_NUMBER_SIGN)
3646 {
3647 ptr++;
3648 while (*ptr != 0)
3649 {
3650 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3651 ptr++;
3652 #ifdef SUPPORT_UTF
3653 if (utf) FORWARDCHAR(ptr);
3654 #endif
3655 }
3656 if (*ptr != 0) continue;
3657
3658 /* Else fall through to handle end of string */
3659 c = 0;
3660 }
3661 }
3662
3663 /* No auto callout for quantifiers. */
3664
3665 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3666 {
3667 previous_callout = code;
3668 code = auto_callout(code, ptr, cd);
3669 }
3670
3671 switch(c)
3672 {
3673 /* ===================================================================*/
3674 case 0: /* The branch terminates at string end */
3675 case CHAR_VERTICAL_LINE: /* or | or ) */
3676 case CHAR_RIGHT_PARENTHESIS:
3677 *firstcharptr = firstchar;
3678 *reqcharptr = reqchar;
3679 *codeptr = code;
3680 *ptrptr = ptr;
3681 if (lengthptr != NULL)
3682 {
3683 if (OFLOW_MAX - *lengthptr < code - last_code)
3684 {
3685 *errorcodeptr = ERR20;
3686 goto FAILED;
3687 }
3688 *lengthptr += (int)(code - last_code); /* To include callout length */
3689 DPRINTF((">> end branch\n"));
3690 }
3691 return TRUE;
3692
3693
3694 /* ===================================================================*/
3695 /* Handle single-character metacharacters. In multiline mode, ^ disables
3696 the setting of any following char as a first character. */
3697
3698 case CHAR_CIRCUMFLEX_ACCENT:
3699 previous = NULL;
3700 if ((options & PCRE_MULTILINE) != 0)
3701 {
3702 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3703 *code++ = OP_CIRCM;
3704 }
3705 else *code++ = OP_CIRC;
3706 break;
3707
3708 case CHAR_DOLLAR_SIGN:
3709 previous = NULL;
3710 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3711 break;
3712
3713 /* There can never be a first char if '.' is first, whatever happens about
3714 repeats. The value of reqchar doesn't change either. */
3715
3716 case CHAR_DOT:
3717 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3718 zerofirstchar = firstchar;
3719 zeroreqchar = reqchar;
3720 previous = code;
3721 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3722 break;
3723
3724
3725 /* ===================================================================*/
3726 /* Character classes. If the included characters are all < 256, we build a
3727 32-byte bitmap of the permitted characters, except in the special case
3728 where there is only one such character. For negated classes, we build the
3729 map as usual, then invert it at the end. However, we use a different opcode
3730 so that data characters > 255 can be handled correctly.
3731
3732 If the class contains characters outside the 0-255 range, a different
3733 opcode is compiled. It may optionally have a bit map for characters < 256,
3734 but those above are are explicitly listed afterwards. A flag byte tells
3735 whether the bitmap is present, and whether this is a negated class or not.
3736
3737 In JavaScript compatibility mode, an isolated ']' causes an error. In
3738 default (Perl) mode, it is treated as a data character. */
3739
3740 case CHAR_RIGHT_SQUARE_BRACKET:
3741 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3742 {
3743 *errorcodeptr = ERR64;
3744 goto FAILED;
3745 }
3746 goto NORMAL_CHAR;
3747
3748 case CHAR_LEFT_SQUARE_BRACKET:
3749 previous = code;
3750
3751 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3752 they are encountered at the top level, so we'll do that too. */
3753
3754 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3755 ptr[1] == CHAR_EQUALS_SIGN) &&
3756 check_posix_syntax(ptr, &tempptr))
3757 {
3758 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3759 goto FAILED;
3760 }
3761
3762 /* If the first character is '^', set the negation flag and skip it. Also,
3763 if the first few characters (either before or after ^) are \Q\E or \E we
3764 skip them too. This makes for compatibility with Perl. */
3765
3766 negate_class = FALSE;
3767 for (;;)
3768 {
3769 c = *(++ptr);
3770 if (c == CHAR_BACKSLASH)
3771 {
3772 if (ptr[1] == CHAR_E)
3773 ptr++;
3774 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3775 ptr += 3;
3776 else
3777 break;
3778 }
3779 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3780 negate_class = TRUE;
3781 else break;
3782 }
3783
3784 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3785 an initial ']' is taken as a data character -- the code below handles
3786 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3787 [^] must match any character, so generate OP_ALLANY. */
3788
3789 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3790 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3791 {
3792 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3793 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3794 zerofirstchar = firstchar;
3795 break;
3796 }
3797
3798 /* If a class contains a negative special such as \S, we need to flip the
3799 negation flag at the end, so that support for characters > 255 works
3800 correctly (they are all included in the class). */
3801
3802 should_flip_negation = FALSE;
3803
3804 /* For optimization purposes, we track some properties of the class.
3805 class_has_8bitchar will be non-zero, if the class contains at least one
3806 < 256 character. class_single_char will be 1 if the class contains only
3807 a single character. */
3808
3809 class_has_8bitchar = 0;
3810 class_single_char = 0;
3811
3812 /* Initialize the 32-char bit map to all zeros. We build the map in a
3813 temporary bit of memory, in case the class contains only 1 character (less
3814 than 256), because in that case the compiled code doesn't use the bit map.
3815 */
3816
3817 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3818
3819 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3820 xclass = FALSE; /* No chars >= 256 */
3821 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3822 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3823 #endif
3824
3825 /* Process characters until ] is reached. By writing this as a "do" it
3826 means that an initial ] is taken as a data character. At the start of the
3827 loop, c contains the first byte of the character. */
3828
3829 if (c != 0) do
3830 {
3831 const pcre_uchar *oldptr;
3832
3833 #ifdef SUPPORT_UTF
3834 if (utf && HAS_EXTRALEN(c))
3835 { /* Braces are required because the */
3836 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3837 }
3838 #endif
3839
3840 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3841 /* In the pre-compile phase, accumulate the length of any extra
3842 data and reset the pointer. This is so that very large classes that
3843 contain a zillion > 255 characters no longer overwrite the work space
3844 (which is on the stack). */
3845
3846 if (lengthptr != NULL)
3847 {
3848 *lengthptr += class_uchardata - class_uchardata_base;
3849 class_uchardata = class_uchardata_base;
3850 }
3851 #endif
3852
3853 /* Inside \Q...\E everything is literal except \E */
3854
3855 if (inescq)
3856 {
3857 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3858 {
3859 inescq = FALSE; /* Reset literal state */
3860 ptr++; /* Skip the 'E' */
3861 continue; /* Carry on with next */
3862 }
3863 goto CHECK_RANGE; /* Could be range if \E follows */
3864 }
3865
3866 /* Handle POSIX class names. Perl allows a negation extension of the
3867 form [:^name:]. A square bracket that doesn't match the syntax is
3868 treated as a literal. We also recognize the POSIX constructions
3869 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3870 5.6 and 5.8 do. */
3871
3872 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3873 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3874 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3875 {
3876 BOOL local_negate = FALSE;
3877 int posix_class, taboffset, tabopt;
3878 register const pcre_uint8 *cbits = cd->cbits;
3879 pcre_uint8 pbits[32];
3880
3881 if (ptr[1] != CHAR_COLON)
3882 {
3883 *errorcodeptr = ERR31;
3884 goto FAILED;
3885 }
3886
3887 ptr += 2;
3888 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3889 {
3890 local_negate = TRUE;
3891 should_flip_negation = TRUE; /* Note negative special */
3892 ptr++;
3893 }
3894
3895 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3896 if (posix_class < 0)
3897 {
3898 *errorcodeptr = ERR30;
3899 goto FAILED;
3900 }
3901
3902 /* If matching is caseless, upper and lower are converted to
3903 alpha. This relies on the fact that the class table starts with
3904 alpha, lower, upper as the first 3 entries. */
3905
3906 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3907 posix_class = 0;
3908
3909 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3910 different escape sequences that use Unicode properties. */
3911
3912 #ifdef SUPPORT_UCP
3913 if ((options & PCRE_UCP) != 0)
3914 {
3915 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3916 if (posix_substitutes[pc] != NULL)
3917 {
3918 nestptr = tempptr + 1;
3919 ptr = posix_substitutes[pc] - 1;
3920 continue;
3921 }
3922 }
3923 #endif
3924 /* In the non-UCP case, we build the bit map for the POSIX class in a
3925 chunk of local store because we may be adding and subtracting from it,
3926 and we don't want to subtract bits that may be in the main map already.
3927 At the end we or the result into the bit map that is being built. */
3928
3929 posix_class *= 3;
3930
3931 /* Copy in the first table (always present) */
3932
3933 memcpy(pbits, cbits + posix_class_maps[posix_class],
3934 32 * sizeof(pcre_uint8));
3935
3936 /* If there is a second table, add or remove it as required. */
3937
3938 taboffset = posix_class_maps[posix_class + 1];
3939 tabopt = posix_class_maps[posix_class + 2];
3940
3941 if (taboffset >= 0)
3942 {
3943 if (tabopt >= 0)
3944 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3945 else
3946 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3947 }
3948
3949 /* Not see if we need to remove any special characters. An option
3950 value of 1 removes vertical space and 2 removes underscore. */
3951
3952 if (tabopt < 0) tabopt = -tabopt;
3953 if (tabopt == 1) pbits[1] &= ~0x3c;
3954 else if (tabopt == 2) pbits[11] &= 0x7f;
3955
3956 /* Add the POSIX table or its complement into the main table that is
3957 being built and we are done. */
3958
3959 if (local_negate)
3960 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3961 else
3962 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3963
3964 ptr = tempptr + 1;
3965 /* Every class contains at least one < 256 characters. */
3966 class_has_8bitchar = 1;
3967 /* Every class contains at least two characters. */
3968 class_single_char = 2;
3969 continue; /* End of POSIX syntax handling */
3970 }
3971
3972 /* Backslash may introduce a single character, or it may introduce one
3973 of the specials, which just set a flag. The sequence \b is a special
3974 case. Inside a class (and only there) it is treated as backspace. We
3975 assume that other escapes have more than one character in them, so
3976 speculatively set both class_has_8bitchar and class_single_char bigger
3977 than one. Unrecognized escapes fall through and are either treated
3978 as literal characters (by default), or are faulted if
3979 PCRE_EXTRA is set. */
3980
3981 if (c == CHAR_BACKSLASH)
3982 {
3983 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3984 if (*errorcodeptr != 0) goto FAILED;
3985
3986 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3987 else if (-c == ESC_N) /* \N is not supported in a class */
3988 {
3989 *errorcodeptr = ERR71;
3990 goto FAILED;
3991 }
3992 else if (-c == ESC_Q) /* Handle start of quoted string */
3993 {
3994 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3995 {
3996 ptr += 2; /* avoid empty string */
3997 }
3998 else inescq = TRUE;
3999 continue;
4000 }
4001 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4002
4003 if (c < 0)
4004 {
4005 register const pcre_uint8 *cbits = cd->cbits;
4006 /* Every class contains at least two < 256 characters. */
4007 class_has_8bitchar++;
4008 /* Every class contains at least two characters. */
4009 class_single_char += 2;
4010
4011 switch (-c)
4012 {
4013 #ifdef SUPPORT_UCP
4014 case ESC_du: /* These are the values given for \d etc */
4015 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4016 case ESC_wu: /* escape sequence with an appropriate \p */
4017 case ESC_WU: /* or \P to test Unicode properties instead */
4018 case ESC_su: /* of the default ASCII testing. */
4019 case ESC_SU:
4020 nestptr = ptr;
4021 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4022 class_has_8bitchar--; /* Undo! */
4023 continue;
4024 #endif
4025 case ESC_d:
4026 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4027 continue;
4028
4029 case ESC_D:
4030 should_flip_negation = TRUE;
4031 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4032 continue;
4033
4034 case ESC_w:
4035 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4036 continue;
4037
4038 case ESC_W:
4039 should_flip_negation = TRUE;
4040 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4041 continue;
4042
4043 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4044 if it was previously set by something earlier in the character
4045 class. */
4046
4047 case ESC_s:
4048 classbits[0] |= cbits[cbit_space];
4049 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4050 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4051 continue;
4052
4053 case ESC_S:
4054 should_flip_negation = TRUE;
4055 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4056 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4057 continue;
4058
4059 case ESC_h:
4060 SETBIT(classbits, 0x09); /* VT */
4061 SETBIT(classbits, 0x20); /* SPACE */
4062 SETBIT(classbits, 0xa0); /* NSBP */
4063 #ifndef COMPILE_PCRE8
4064 xclass = TRUE;
4065 *class_uchardata++ = XCL_SINGLE;
4066 *class_uchardata++ = 0x1680;
4067 *class_uchardata++ = XCL_SINGLE;
4068 *class_uchardata++ = 0x180e;
4069 *class_uchardata++ = XCL_RANGE;
4070 *class_uchardata++ = 0x2000;
4071 *class_uchardata++ = 0x200a;
4072 *class_uchardata++ = XCL_SINGLE;
4073 *class_uchardata++ = 0x202f;
4074 *class_uchardata++ = XCL_SINGLE;
4075 *class_uchardata++ = 0x205f;
4076 *class_uchardata++ = XCL_SINGLE;
4077 *class_uchardata++ = 0x3000;
4078 #elif defined SUPPORT_UTF
4079 if (utf)
4080 {
4081 xclass = TRUE;
4082 *class_uchardata++ = XCL_SINGLE;
4083 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4084 *class_uchardata++ = XCL_SINGLE;
4085 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4086 *class_uchardata++ = XCL_RANGE;
4087 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4088 class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
4089 *class_uchardata++ = XCL_SINGLE;
4090 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4091 *class_uchardata++ = XCL_SINGLE;
4092 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4093 *class_uchardata++ = XCL_SINGLE;
4094 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4095 }
4096 #endif
4097 continue;
4098
4099 case ESC_H:
4100 for (c = 0; c < 32; c++)
4101 {
4102 int x = 0xff;
4103 switch (c)
4104 {
4105 case 0x09/8: x ^= 1 << (0x09%8); break;
4106 case 0x20/8: x ^= 1 << (0x20%8); break;
4107 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4108 default: break;
4109 }
4110 classbits[c] |= x;
4111 }
4112 #ifndef COMPILE_PCRE8
4113 xclass = TRUE;
4114 *class_uchardata++ = XCL_RANGE;
4115 *class_uchardata++ = 0x0100;
4116 *class_uchardata++ = 0x167f;
4117 *class_uchardata++ = XCL_RANGE;
4118 *class_uchardata++ = 0x1681;
4119 *class_uchardata++ = 0x180d;
4120 *class_uchardata++ = XCL_RANGE;
4121 *class_uchardata++ = 0x180f;
4122 *class_uchardata++ = 0x1fff;
4123 *class_uchardata++ = XCL_RANGE;
4124 *class_uchardata++ = 0x200b;
4125 *class_uchardata++ = 0x202e;
4126 *class_uchardata++ = XCL_RANGE;
4127 *class_uchardata++ = 0x2030;
4128 *class_uchardata++ = 0x205e;
4129 *class_uchardata++ = XCL_RANGE;
4130 *class_uchardata++ = 0x2060;
4131 *class_uchardata++ = 0x2fff;
4132 *class_uchardata++ = XCL_RANGE;
4133 *class_uchardata++ = 0x3001;
4134 #ifdef SUPPORT_UTF
4135 if (utf)
4136 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4137 else
4138 #endif
4139 *class_uchardata++ = 0xffff;
4140 #elif defined SUPPORT_UTF
4141 if (utf)
4142 {
4143 xclass = TRUE;
4144 *class_uchardata++ = XCL_RANGE;
4145 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4146 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4147 *class_uchardata++ = XCL_RANGE;
4148 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4149 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4150 *class_uchardata++ = XCL_RANGE;
4151 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4152 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4153 *class_uchardata++ = XCL_RANGE;
4154 class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
4155 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4156 *class_uchardata++ = XCL_RANGE;
4157 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4158 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4159 *class_uchardata++ = XCL_RANGE;
4160 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4161 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4162 *class_uchardata++ = XCL_RANGE;
4163 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4164 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4165 }
4166 #endif
4167 continue;
4168
4169 case ESC_v:
4170 SETBIT(classbits, 0x0a); /* LF */
4171 SETBIT(classbits, 0x0b); /* VT */
4172 SETBIT(classbits, 0x0c); /* FF */
4173 SETBIT(classbits, 0x0d); /* CR */
4174 SETBIT(classbits, 0x85); /* NEL */
4175 #ifndef COMPILE_PCRE8
4176 xclass = TRUE;
4177 *class_uchardata++ = XCL_RANGE;
4178 *class_uchardata++ = 0x2028;
4179 *class_uchardata++ = 0x2029;
4180 #elif defined SUPPORT_UTF
4181 if (utf)
4182 {
4183 xclass = TRUE;
4184 *class_uchardata++ = XCL_RANGE;
4185 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4186 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4187 }
4188 #endif
4189 continue;
4190
4191 case ESC_V:
4192 for (c = 0; c < 32; c++)
4193 {
4194 int x = 0xff;
4195 switch (c)
4196 {
4197 case 0x0a/8: x ^= 1 << (0x0a%8);
4198 x ^= 1 << (0x0b%8);
4199 x ^= 1 << (0x0c%8);
4200 x ^= 1 << (0x0d%8);
4201 break;
4202 case 0x85/8: x ^= 1 << (0x85%8); break;
4203 default: break;
4204 }
4205 classbits[c] |= x;
4206 }
4207
4208 #ifndef COMPILE_PCRE8
4209 xclass = TRUE;
4210 *class_uchardata++ = XCL_RANGE;
4211 *class_uchardata++ = 0x0100;
4212 *class_uchardata++ = 0x2027;
4213 *class_uchardata++ = XCL_RANGE;
4214 *class_uchardata++ = 0x202a;
4215 #ifdef SUPPORT_UTF
4216 if (utf)
4217 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4218 else
4219 #endif
4220 *class_uchardata++ = 0xffff;
4221 #elif defined SUPPORT_UTF
4222 if (utf)
4223 {
4224 xclass = TRUE;
4225 *class_uchardata++ = XCL_RANGE;
4226 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4227 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4228 *class_uchardata++ = XCL_RANGE;
4229 class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
4230 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4231 }
4232 #endif
4233 continue;
4234
4235 #ifdef SUPPORT_UCP
4236 case ESC_p:
4237 case ESC_P:
4238 {
4239 BOOL negated;
4240 int pdata;
4241 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4242 if (ptype < 0) goto FAILED;
4243 xclass = TRUE;
4244 *class_uchardata++ = ((-c == ESC_p) != negated)?
4245 XCL_PROP : XCL_NOTPROP;
4246 *class_uchardata++ = ptype;
4247 *class_uchardata++ = pdata;
4248 class_has_8bitchar--; /* Undo! */
4249 continue;
4250 }
4251 #endif
4252 /* Unrecognized escapes are faulted if PCRE is running in its
4253 strict mode. By default, for compatibility with Perl, they are
4254 treated as literals. */
4255
4256 default:
4257 if ((options & PCRE_EXTRA) != 0)
4258 {
4259 *errorcodeptr = ERR7;
4260 goto FAILED;
4261 }
4262 class_has_8bitchar--; /* Undo the speculative increase. */
4263 class_single_char -= 2; /* Undo the speculative increase. */
4264 c = *ptr; /* Get the final character and fall through */
4265 break;
4266 }
4267 }
4268
4269 /* Fall through if we have a single character (c >= 0). This may be
4270 greater than 256. */
4271
4272 } /* End of backslash handling */
4273
4274 /* A single character may be followed by '-' to form a range. However,
4275 Perl does not permit ']' to be the end of the range. A '-' character
4276 at the end is treated as a literal. Perl ignores orphaned \E sequences
4277 entirely. The code for handling \Q and \E is messy. */
4278
4279 CHECK_RANGE:
4280 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4281 {
4282 inescq = FALSE;
4283 ptr += 2;
4284 }
4285
4286 oldptr = ptr;
4287
4288 /* Remember \r or \n */
4289
4290 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4291
4292 /* Check for range */
4293
4294 if (!inescq && ptr[1] == CHAR_MINUS)
4295 {
4296 int d;
4297 ptr += 2;
4298 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4299
4300 /* If we hit \Q (not followed by \E) at this point, go into escaped
4301 mode. */
4302
4303 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4304 {
4305 ptr += 2;
4306 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4307 { ptr += 2; continue; }
4308 inescq = TRUE;
4309 break;
4310 }
4311
4312 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4313 {
4314 ptr = oldptr;
4315 goto LONE_SINGLE_CHARACTER;
4316 }
4317
4318 #ifdef SUPPORT_UTF
4319 if (utf)
4320 { /* Braces are required because the */
4321 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4322 }
4323 else
4324 #endif
4325 d = *ptr; /* Not UTF-8 mode */
4326
4327 /* The second part of a range can be a single-character escape, but
4328 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4329 in such circumstances. */
4330
4331 if (!inescq && d == CHAR_BACKSLASH)
4332 {
4333 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4334 if (*errorcodeptr != 0) goto FAILED;
4335
4336 /* \b is backspace; any other special means the '-' was literal */
4337
4338 if (d < 0)
4339 {
4340 if (d == -ESC_b) d = CHAR_BS; else
4341 {
4342 ptr = oldptr;
4343 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4344 }
4345 }
4346 }
4347
4348 /* Check that the two values are in the correct order. Optimize
4349 one-character ranges */
4350
4351 if (d < c)
4352 {
4353 *errorcodeptr = ERR8;
4354 goto FAILED;
4355 }
4356
4357 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4358
4359 /* Remember \r or \n */
4360
4361 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4362
4363 /* Since we found a character range, single character optimizations
4364 cannot be done anymore. */
4365 class_single_char = 2;
4366
4367 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4368 matching, we have to use an XCLASS with extra data items. Caseless
4369 matching for characters > 127 is available only if UCP support is
4370 available. */
4371
4372 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4373 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4374 #elif defined SUPPORT_UTF
4375 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4376 #elif !(defined COMPILE_PCRE8)
4377 if (d > 255)
4378 #endif
4379 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4380 {
4381 xclass = TRUE;
4382
4383 /* With UCP support, we can find the other case equivalents of
4384 the relevant characters. There may be several ranges. Optimize how
4385 they fit with the basic range. */
4386
4387 #ifdef SUPPORT_UCP
4388 #ifndef COMPILE_PCRE8
4389 if (utf && (options & PCRE_CASELESS) != 0)
4390 #else
4391 if ((options & PCRE_CASELESS) != 0)
4392 #endif
4393 {
4394 unsigned int occ, ocd;
4395 unsigned int cc = c;
4396 unsigned int origd = d;
4397 while (get_othercase_range(&cc, origd, &occ, &ocd))
4398 {
4399 if (occ >= (unsigned int)c &&
4400 ocd <= (unsigned int)d)
4401 continue; /* Skip embedded ranges */
4402
4403 if (occ < (unsigned int)c &&
4404 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4405 { /* if there is overlap, */
4406 c = occ; /* noting that if occ < c */
4407 continue; /* we can't have ocd > d */
4408 } /* because a subrange is */
4409 if (ocd > (unsigned int)d &&
4410 occ <= (unsigned int)d + 1) /* always shorter than */
4411 { /* the basic range. */
4412 d = ocd;
4413 continue;
4414 }
4415
4416 if (occ == ocd)
4417 {
4418 *class_uchardata++ = XCL_SINGLE;
4419 }
4420 else
4421 {
4422 *class_uchardata++ = XCL_RANGE;
4423 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4424 }
4425 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4426 }
4427 }
4428 #endif /* SUPPORT_UCP */
4429
4430 /* Now record the original range, possibly modified for UCP caseless
4431 overlapping ranges. */
4432
4433 *class_uchardata++ = XCL_RANGE;
4434 #ifdef SUPPORT_UTF
4435 #ifndef COMPILE_PCRE8
4436 if (utf)
4437 {
4438 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4439 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4440 }
4441 else
4442 {
4443 *class_uchardata++ = c;
4444 *class_uchardata++ = d;
4445 }
4446 #else
4447 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4448 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4449 #endif
4450 #else /* SUPPORT_UTF */
4451 *class_uchardata++ = c;
4452 *class_uchardata++ = d;
4453 #endif /* SUPPORT_UTF */
4454
4455 /* With UCP support, we are done. Without UCP support, there is no
4456 caseless matching for UTF characters > 127; we can use the bit map
4457 for the smaller ones. As for 16 bit characters without UTF, we
4458 can still use */
4459
4460 #ifdef SUPPORT_UCP
4461 #ifndef COMPILE_PCRE8
4462 if (utf)
4463 #endif
4464 continue; /* With next character in the class */
4465 #endif /* SUPPORT_UCP */
4466
4467 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4468 if (utf)
4469 {
4470 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4471 /* Adjust upper limit and fall through to set up the map */
4472 d = 127;
4473 }
4474 else
4475 {
4476 if (c > 255) continue;
4477 /* Adjust upper limit and fall through to set up the map */
4478 d = 255;
4479 }
4480 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4481 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4482 /* Adjust upper limit and fall through to set up the map */
4483 d = 127;
4484 #else
4485 if (c > 255) continue;
4486 /* Adjust upper limit and fall through to set up the map */
4487 d = 255;
4488 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4489 }
4490 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4491
4492 /* We use the bit map for 8 bit mode, or when the characters fall
4493 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4494
4495 class_has_8bitchar = 1;
4496
4497 /* We can save a bit of time by skipping this in the pre-compile. */
4498
4499 if (lengthptr == NULL) for (; c <= d; c++)
4500 {
4501 classbits[c/8] |= (1 << (c&7));
4502 if ((options & PCRE_CASELESS) != 0)
4503 {
4504 int uc = cd->fcc[c]; /* flip case */
4505 classbits[uc/8] |= (1 << (uc&7));
4506 }
4507 }
4508
4509 continue; /* Go get the next char in the class */
4510 }
4511
4512 /* Handle a lone single character - we can get here for a normal
4513 non-escape char, or after \ that introduces a single character or for an
4514 apparent range that isn't. */
4515
4516 LONE_SINGLE_CHARACTER:
4517
4518 /* Only the value of 1 matters for class_single_char. */
4519
4520 if (class_single_char < 2) class_single_char++;
4521
4522 /* If class_charcount is 1, we saw precisely one character. As long as
4523 there was no use of \p or \P, in other words, no use of any XCLASS
4524 features, we can optimize.
4525
4526 The optimization throws away the bit map. We turn the item into a
4527 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4528 In the positive case, it can cause firstchar to be set. Otherwise, there
4529 can be no first char if this item is first, whatever repeat count may
4530 follow. In the case of reqchar, save the previous value for reinstating. */
4531
4532 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4533 {
4534 ptr++;
4535 zeroreqchar = reqchar;
4536
4537 if (negate_class)
4538 {
4539 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4540 zerofirstchar = firstchar;
4541 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4542 #ifdef SUPPORT_UTF
4543 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4544 code += PRIV(ord2utf)(c, code);
4545 else
4546 #endif
4547 *code++ = c;
4548 goto NOT_CHAR;
4549 }
4550
4551 /* For a single, positive character, get the value into mcbuffer, and
4552 then we can handle this with the normal one-character code. */
4553
4554 #ifdef SUPPORT_UTF
4555 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4556 mclength = PRIV(ord2utf)(c, mcbuffer);
4557 else
4558 #endif
4559 {
4560 mcbuffer[0] = c;
4561 mclength = 1;
4562 }
4563 goto ONE_CHAR;
4564 } /* End of 1-char optimization */
4565
4566 /* Handle a character that cannot go in the bit map. */
4567
4568 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4569 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4570 #elif defined SUPPORT_UTF
4571 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4572 #elif !(defined COMPILE_PCRE8)
4573 if (c > 255)
4574 #endif
4575
4576 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4577 {
4578 xclass = TRUE;
4579 *class_uchardata++ = XCL_SINGLE;
4580 #ifdef SUPPORT_UTF
4581 #ifndef COMPILE_PCRE8
4582 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4583 if (!utf)
4584 *class_uchardata++ = c;
4585 else
4586 #endif
4587 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4588 #else /* SUPPORT_UTF */
4589 *class_uchardata++ = c;
4590 #endif /* SUPPORT_UTF */
4591
4592 #ifdef SUPPORT_UCP
4593 #ifdef COMPILE_PCRE8
4594 if ((options & PCRE_CASELESS) != 0)
4595 #else
4596 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4597 if (utf && (options & PCRE_CASELESS) != 0)
4598 #endif
4599 {
4600 unsigned int othercase;
4601 if ((int)(othercase = UCD_OTHERCASE(c)) != c)
4602 {
4603 *class_uchardata++ = XCL_SINGLE;
4604 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4605 }
4606 }
4607 #endif /* SUPPORT_UCP */
4608
4609 }
4610 else
4611 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4612
4613 /* Handle a single-byte character */
4614 {
4615 class_has_8bitchar = 1;
4616 classbits[c/8] |= (1 << (c&7));
4617 if ((options & PCRE_CASELESS) != 0)
4618 {
4619 c = cd->fcc[c]; /* flip case */
4620 classbits[c/8] |= (1 << (c&7));
4621 }
4622 }
4623 }
4624
4625 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4626 If we are at the end of an internal nested string, revert to the outer
4627 string. */
4628
4629 while (((c = *(++ptr)) != 0 ||
4630 (nestptr != NULL &&
4631 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4632 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4633
4634 /* Check for missing terminating ']' */
4635
4636 if (c == 0)
4637 {
4638 *errorcodeptr = ERR6;
4639 goto FAILED;
4640 }
4641
4642 /* If this is the first thing in the branch, there can be no first char
4643 setting, whatever the repeat count. Any reqchar setting must remain
4644 unchanged after any kind of repeat. */
4645
4646 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4647 zerofirstchar = firstchar;
4648 zeroreqchar = reqchar;
4649
4650 /* If there are characters with values > 255, we have to compile an
4651 extended class, with its own opcode, unless there was a negated special
4652 such as \S in the class, and PCRE_UCP is not set, because in that case all
4653 characters > 255 are in the class, so any that were explicitly given as
4654 well can be ignored. If (when there are explicit characters > 255 that must
4655 be listed) there are no characters < 256, we can omit the bitmap in the
4656 actual compiled code. */
4657
4658 #ifdef SUPPORT_UTF
4659 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4660 #elif !defined COMPILE_PCRE8
4661 if (xclass && !should_flip_negation)
4662 #endif
4663 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4664 {
4665 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4666 *code++ = OP_XCLASS;
4667 code += LINK_SIZE;
4668 *code = negate_class? XCL_NOT:0;
4669
4670 /* If the map is required, move up the extra data to make room for it;
4671 otherwise just move the code pointer to the end of the extra data. */
4672
4673 if (class_has_8bitchar > 0)
4674 {
4675 *code++ |= XCL_MAP;
4676 memmove(code + (32 / sizeof(pcre_uchar)), code,
4677 IN_UCHARS(class_uchardata - code));
4678 memcpy(code, classbits, 32);
4679 code = class_uchardata + (32 / sizeof(pcre_uchar));
4680 }
4681 else code = class_uchardata;
4682
4683 /* Now fill in the complete length of the item */
4684
4685 PUT(previous, 1, (int)(code - previous));
4686 break; /* End of class handling */
4687 }
4688 #endif
4689
4690 /* If there are no characters > 255, or they are all to be included or
4691 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4692 whole class was negated and whether there were negative specials such as \S
4693 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4694 negating it if necessary. */
4695
4696 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4697 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4698 {
4699 if (negate_class)
4700 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4701 memcpy(code, classbits, 32);
4702 }
4703 code += 32 / sizeof(pcre_uchar);
4704 NOT_CHAR:
4705 break;
4706
4707
4708 /* ===================================================================*/
4709 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4710 has been tested above. */
4711
4712 case CHAR_LEFT_CURLY_BRACKET:
4713 if (!is_quantifier) goto NORMAL_CHAR;
4714 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4715 if (*errorcodeptr != 0) goto FAILED;
4716 goto REPEAT;
4717
4718 case CHAR_ASTERISK:
4719 repeat_min = 0;
4720 repeat_max = -1;
4721 goto REPEAT;
4722
4723 case CHAR_PLUS:
4724 repeat_min = 1;
4725 repeat_max = -1;
4726 goto REPEAT;
4727
4728 case CHAR_QUESTION_MARK:
4729 repeat_min = 0;
4730 repeat_max = 1;
4731
4732 REPEAT:
4733 if (previous == NULL)
4734 {
4735 *errorcodeptr = ERR9;
4736 goto FAILED;
4737 }
4738
4739 if (repeat_min == 0)
4740 {
4741 firstchar = zerofirstchar; /* Adjust for zero repeat */
4742 reqchar = zeroreqchar; /* Ditto */
4743 }
4744
4745 /* Remember whether this is a variable length repeat */
4746
4747 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4748
4749 op_type = 0; /* Default single-char op codes */
4750 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4751
4752 /* Save start of previous item, in case we have to move it up in order to
4753 insert something before it. */
4754
4755 tempcode = previous;
4756
4757 /* If the next character is '+', we have a possessive quantifier. This
4758 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4759 If the next character is '?' this is a minimizing repeat, by default,
4760 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4761 repeat type to the non-default. */
4762
4763 if (ptr[1] == CHAR_PLUS)
4764 {
4765 repeat_type = 0; /* Force greedy */
4766 possessive_quantifier = TRUE;
4767 ptr++;
4768 }
4769 else if (ptr[1] == CHAR_QUESTION_MARK)
4770 {
4771 repeat_type = greedy_non_default;
4772 ptr++;
4773 }
4774 else repeat_type = greedy_default;
4775
4776 /* If previous was a recursion call, wrap it in atomic brackets so that
4777 previous becomes the atomic group. All recursions were so wrapped in the
4778 past, but it no longer happens for non-repeated recursions. In fact, the
4779 repeated ones could be re-implemented independently so as not to need this,
4780 but for the moment we rely on the code for repeating groups. */
4781
4782 if (*previous == OP_RECURSE)
4783 {
4784 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4785 *previous = OP_ONCE;
4786 PUT(previous, 1, 2 + 2*LINK_SIZE);
4787 previous[2 + 2*LINK_SIZE] = OP_KET;
4788 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4789 code += 2 + 2 * LINK_SIZE;
4790 length_prevgroup = 3 + 3*LINK_SIZE;
4791
4792 /* When actually compiling, we need to check whether this was a forward
4793 reference, and if so, adjust the offset. */
4794
4795 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4796 {
4797 int offset = GET(cd->hwm, -LINK_SIZE);
4798 if (offset == previous + 1 - cd->start_code)
4799 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4800 }
4801 }
4802
4803 /* Now handle repetition for the different types of item. */
4804
4805 /* If previous was a character or negated character match, abolish the item
4806 and generate a repeat item instead. If a char item has a minimum of more
4807 than one, ensure that it is set in reqchar - it might not be if a sequence
4808 such as x{3} is the first thing in a branch because the x will have gone
4809 into firstchar instead. */
4810
4811 if (*previous == OP_CHAR || *previous == OP_CHARI
4812 || *previous == OP_NOT || *previous == OP_NOTI)
4813 {
4814 switch (*previous)
4815 {
4816 default: /* Make compiler happy. */
4817 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4818 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4819 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4820 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4821 }
4822
4823 /* Deal with UTF characters that take up more than one character. It's
4824 easier to write this out separately than try to macrify it. Use c to
4825 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4826 it's a length rather than a small character. */
4827
4828 #ifdef SUPPORT_UTF
4829 if (utf && NOT_FIRSTCHAR(code[-1]))
4830 {
4831 pcre_uchar *lastchar = code - 1;
4832 BACKCHAR(lastchar);
4833 c = (int)(code - lastchar); /* Length of UTF-8 character */
4834 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4835 c |= UTF_LENGTH; /* Flag c as a length */
4836 }
4837 else
4838 #endif /* SUPPORT_UTF */
4839
4840 /* Handle the case of a single charater - either with no UTF support, or
4841 with UTF disabled, or for a single character UTF character. */
4842 {
4843 c = code[-1];
4844 if (*previous <= OP_CHARI && repeat_min > 1)
4845 reqchar = c | req_caseopt | cd->req_varyopt;
4846 }
4847
4848 /* If the repetition is unlimited, it pays to see if the next thing on
4849 the line is something that cannot possibly match this character. If so,
4850 automatically possessifying this item gains some performance in the case
4851 where the match fails. */
4852
4853 if (!possessive_quantifier &&
4854 repeat_max < 0 &&
4855 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4856 {
4857 repeat_type = 0; /* Force greedy */
4858 possessive_quantifier = TRUE;
4859 }
4860
4861 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4862 }
4863
4864 /* If previous was a character type match (\d or similar), abolish it and
4865 create a suitable repeat item. The code is shared with single-character
4866 repeats by setting op_type to add a suitable offset into repeat_type. Note
4867 the the Unicode property types will be present only when SUPPORT_UCP is
4868 defined, but we don't wrap the little bits of code here because it just
4869 makes it horribly messy. */
4870
4871 else if (*previous < OP_EODN)
4872 {
4873 pcre_uchar *oldcode;
4874 int prop_type, prop_value;
4875 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4876 c = *previous;
4877
4878 if (!possessive_quantifier &&
4879 repeat_max < 0 &&
4880 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4881 {
4882 repeat_type = 0; /* Force greedy */
4883 possessive_quantifier = TRUE;
4884 }
4885
4886 OUTPUT_SINGLE_REPEAT:
4887 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4888 {
4889 prop_type = previous[1];
4890 prop_value = previous[2];
4891 }
4892 else prop_type = prop_value = -1;
4893
4894 oldcode = code;
4895 code = previous; /* Usually overwrite previous item */
4896
4897 /* If the maximum is zero then the minimum must also be zero; Perl allows
4898 this case, so we do too - by simply omitting the item altogether. */
4899
4900 if (repeat_max == 0) goto END_REPEAT;
4901
4902 /*--------------------------------------------------------------------*/
4903 /* This code is obsolete from release 8.00; the restriction was finally
4904 removed: */
4905
4906 /* All real repeats make it impossible to handle partial matching (maybe
4907 one day we will be able to remove this restriction). */
4908
4909 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4910 /*--------------------------------------------------------------------*/
4911
4912 /* Combine the op_type with the repeat_type */
4913
4914 repeat_type += op_type;
4915
4916 /* A minimum of zero is handled either as the special case * or ?, or as
4917 an UPTO, with the maximum given. */
4918
4919 if (repeat_min == 0)
4920 {
4921 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4922 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4923 else
4924 {
4925 *code++ = OP_UPTO + repeat_type;
4926 PUT2INC(code, 0, repeat_max);
4927 }
4928 }
4929
4930 /* A repeat minimum of 1 is optimized into some special cases. If the
4931 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4932 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4933 one less than the maximum. */
4934
4935 else if (repeat_min == 1)
4936 {
4937 if (repeat_max == -1)
4938 *code++ = OP_PLUS + repeat_type;
4939 else
4940 {
4941 code = oldcode; /* leave previous item in place */
4942 if (repeat_max == 1) goto END_REPEAT;
4943 *code++ = OP_UPTO + repeat_type;
4944 PUT2INC(code, 0, repeat_max - 1);
4945 }
4946 }
4947
4948 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4949 handled as an EXACT followed by an UPTO. */
4950
4951 else
4952 {
4953 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4954 PUT2INC(code, 0, repeat_min);
4955
4956 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4957 we have to insert the character for the previous code. For a repeated
4958 Unicode property match, there are two extra bytes that define the
4959 required property. In UTF-8 mode, long characters have their length in
4960 c, with the UTF_LENGTH bit as a flag. */
4961
4962 if (repeat_max < 0)
4963 {
4964 #ifdef SUPPORT_UTF
4965 if (utf && (c & UTF_LENGTH) != 0)
4966 {
4967 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4968 code += c & 7;
4969 }
4970 else
4971 #endif
4972 {
4973 *code++ = c;
4974 if (prop_type >= 0)
4975 {
4976 *code++ = prop_type;
4977 *code++ = prop_value;
4978 }
4979 }
4980 *code++ = OP_STAR + repeat_type;
4981 }
4982
4983 /* Else insert an UPTO if the max is greater than the min, again
4984 preceded by the character, for the previously inserted code. If the
4985 UPTO is just for 1 instance, we can use QUERY instead. */
4986
4987 else if (repeat_max != repeat_min)
4988 {
4989 #ifdef SUPPORT_UTF
4990 if (utf && (c & UTF_LENGTH) != 0)
4991 {
4992 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4993 code += c & 7;
4994 }
4995 else
4996 #endif
4997 *code++ = c;
4998 if (prop_type >= 0)
4999 {
5000 *code++ = prop_type;
5001 *code++ = prop_value;
5002 }
5003 repeat_max -= repeat_min;
5004
5005 if (repeat_max == 1)
5006 {
5007 *code++ = OP_QUERY + repeat_type;
5008 }
5009 else
5010 {
5011 *code++ = OP_UPTO + repeat_type;
5012 PUT2INC(code, 0, repeat_max);
5013 }
5014 }
5015 }
5016
5017 /* The character or character type itself comes last in all cases. */
5018
5019 #ifdef SUPPORT_UTF
5020 if (utf && (c & UTF_LENGTH) != 0)
5021 {
5022 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5023 code += c & 7;
5024 }
5025 else
5026 #endif
5027 *code++ = c;
5028
5029 /* For a repeated Unicode property match, there are two extra bytes that
5030 define the required property. */
5031
5032 #ifdef SUPPORT_UCP
5033 if (prop_type >= 0)
5034 {
5035 *code++ = prop_type;
5036 *code++ = prop_value;
5037 }
5038 #endif
5039 }
5040
5041 /* If previous was a character class or a back reference, we put the repeat
5042 stuff after it, but just skip the item if the repeat was {0,0}. */
5043
5044 else if (*previous == OP_CLASS ||
5045 *previous == OP_NCLASS ||
5046 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5047 *previous == OP_XCLASS ||
5048 #endif
5049 *previous == OP_REF ||
5050 *previous == OP_REFI)
5051 {
5052 if (repeat_max == 0)
5053 {
5054 code = previous;
5055 goto END_REPEAT;
5056 }
5057
5058 /*--------------------------------------------------------------------*/
5059 /* This code is obsolete from release 8.00; the restriction was finally
5060 removed: */
5061
5062 /* All real repeats make it impossible to handle partial matching (maybe
5063 one day we will be able to remove this restriction). */
5064
5065 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5066 /*--------------------------------------------------------------------*/
5067
5068 if (repeat_min == 0 && repeat_max == -1)
5069 *code++ = OP_CRSTAR + repeat_type;
5070 else if (repeat_min == 1 && repeat_max == -1)
5071 *code++ = OP_CRPLUS + repeat_type;
5072 else if (repeat_min == 0 && repeat_max == 1)
5073 *code++ = OP_CRQUERY + repeat_type;
5074 else
5075 {
5076 *code++ = OP_CRRANGE + repeat_type;
5077 PUT2INC(code, 0, repeat_min);
5078 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5079 PUT2INC(code, 0, repeat_max);
5080 }
5081 }
5082
5083 /* If previous was a bracket group, we may have to replicate it in certain
5084 cases. Note that at this point we can encounter only the "basic" bracket
5085 opcodes such as BRA and CBRA, as this is the place where they get converted
5086 into the more special varieties such as BRAPOS and SBRA. A test for >=
5087 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5088 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5089 repetition of assertions, but now it does, for Perl compatibility. */
5090
5091 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5092 {
5093 register int i;
5094 int len = (int)(code - previous);
5095 pcre_uchar *bralink = NULL;
5096 pcre_uchar *brazeroptr = NULL;
5097
5098 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5099 we just ignore the repeat. */
5100
5101 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5102 goto END_REPEAT;
5103
5104 /* There is no sense in actually repeating assertions. The only potential
5105 use of repetition is in cases when the assertion is optional. Therefore,
5106 if the minimum is greater than zero, just ignore the repeat. If the
5107 maximum is not not zero or one, set it to 1. */
5108
5109 if (*previous < OP_ONCE) /* Assertion */
5110 {
5111 if (repeat_min > 0) goto END_REPEAT;
5112 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5113 }
5114
5115 /* The case of a zero minimum is special because of the need to stick
5116 OP_BRAZERO in front of it, and because the group appears once in the
5117 data, whereas in other cases it appears the minimum number of times. For
5118 this reason, it is simplest to treat this case separately, as otherwise
5119 the code gets far too messy. There are several special subcases when the
5120 minimum is zero. */
5121
5122 if (repeat_min == 0)
5123 {
5124 /* If the maximum is also zero, we used to just omit the group from the
5125 output altogether, like this:
5126
5127 ** if (repeat_max == 0)
5128 ** {
5129 ** code = previous;
5130 ** goto END_REPEAT;
5131 ** }
5132
5133 However, that fails when a group or a subgroup within it is referenced
5134 as a subroutine from elsewhere in the pattern, so now we stick in
5135 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5136 don't have a list of which groups are referenced, we cannot do this
5137 selectively.
5138
5139 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5140 and do no more at this point. However, we do need to adjust any
5141 OP_RECURSE calls inside the group that refer to the group itself or any
5142 internal or forward referenced group, because the offset is from the
5143 start of the whole regex. Temporarily terminate the pattern while doing
5144 this. */
5145
5146 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5147 {
5148 *code = OP_END;
5149 adjust_recurse(previous, 1, utf, cd, save_hwm);
5150 memmove(previous + 1, previous, IN_UCHARS(len));
5151 code++;
5152 if (repeat_max == 0)
5153 {
5154 *previous++ = OP_SKIPZERO;
5155 goto END_REPEAT;
5156 }
5157 brazeroptr = previous; /* Save for possessive optimizing */
5158 *previous++ = OP_BRAZERO + repeat_type;
5159 }
5160
5161 /* If the maximum is greater than 1 and limited, we have to replicate
5162 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5163 The first one has to be handled carefully because it's the original
5164 copy, which has to be moved up. The remainder can be handled by code
5165 that is common with the non-zero minimum case below. We have to
5166 adjust the value or repeat_max, since one less copy is required. Once
5167 again, we may have to adjust any OP_RECURSE calls inside the group. */
5168
5169 else
5170 {
5171 int offset;
5172 *code = OP_END;
5173 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5174 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5175 code += 2 + LINK_SIZE;
5176 *previous++ = OP_BRAZERO + repeat_type;
5177 *previous++ = OP_BRA;
5178
5179 /* We chain together the bracket offset fields that have to be
5180 filled in later when the ends of the brackets are reached. */
5181
5182 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5183 bralink = previous;
5184 PUTINC(previous, 0, offset);
5185 }
5186
5187 repeat_max--;
5188 }
5189
5190 /* If the minimum is greater than zero, replicate the group as many
5191 times as necessary, and adjust the maximum to the number of subsequent
5192 copies that we need. If we set a first char from the group, and didn't
5193 set a required char, copy the latter from the former. If there are any
5194 forward reference subroutine calls in the group, there will be entries on
5195 the workspace list; replicate these with an appropriate increment. */
5196
5197 else
5198 {
5199 if (repeat_min > 1)
5200 {
5201 /* In the pre-compile phase, we don't actually do the replication. We
5202 just adjust the length as if we had. Do some paranoid checks for
5203 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5204 integer type when available, otherwise double. */
5205
5206 if (lengthptr != NULL)
5207 {
5208 int delta = (repeat_min - 1)*length_prevgroup;
5209 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5210 (INT64_OR_DOUBLE)length_prevgroup >
5211 (INT64_OR_DOUBLE)INT_MAX ||
5212 OFLOW_MAX - *lengthptr < delta)
5213 {
5214 *errorcodeptr = ERR20;
5215 goto FAILED;
5216 }
5217 *lengthptr += delta;
5218 }
5219
5220 /* This is compiling for real. If there is a set first byte for
5221 the group, and we have not yet set a "required byte", set it. Make
5222 sure there is enough workspace for copying forward references before
5223 doing the copy. */
5224
5225 else
5226 {
5227 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5228
5229 for (i = 1; i < repeat_min; i++)
5230 {
5231 pcre_uchar *hc;
5232 pcre_uchar *this_hwm = cd->hwm;
5233 memcpy(code, previous, IN_UCHARS(len));
5234
5235 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5236 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5237 {
5238 int save_offset = save_hwm - cd->start_workspace;
5239 int this_offset = this_hwm - cd->start_workspace;
5240 *errorcodeptr = expand_workspace(cd);
5241 if (*errorcodeptr != 0) goto FAILED;
5242 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5243 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5244 }
5245
5246 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5247 {
5248 PUT(cd->hwm, 0, GET(hc, 0) + len);
5249 cd->hwm += LINK_SIZE;
5250 }
5251 save_hwm = this_hwm;
5252 code += len;
5253 }
5254 }
5255 }
5256
5257 if (repeat_max > 0) repeat_max -= repeat_min;
5258 }
5259
5260 /* This code is common to both the zero and non-zero minimum cases. If
5261 the maximum is limited, it replicates the group in a nested fashion,
5262 remembering the bracket starts on a stack. In the case of a zero minimum,
5263 the first one was set up above. In all cases the repeat_max now specifies
5264 the number of additional copies needed. Again, we must remember to
5265 replicate entries on the forward reference list. */
5266
5267 if (repeat_max >= 0)
5268 {
5269 /* In the pre-compile phase, we don't actually do the replication. We
5270 just adjust the length as if we had. For each repetition we must add 1
5271 to the length for BRAZERO and for all but the last repetition we must
5272 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5273 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5274 a 64-bit integer type when available, otherwise double. */
5275
5276 if (lengthptr != NULL && repeat_max > 0)
5277 {
5278 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5279 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5280 if ((INT64_OR_DOUBLE)repeat_max *
5281 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5282 > (INT64_OR_DOUBLE)INT_MAX ||
5283 OFLOW_MAX - *lengthptr < delta)
5284 {
5285 *errorcodeptr = ERR20;
5286 goto FAILED;
5287 }
5288 *lengthptr += delta;
5289 }
5290
5291 /* This is compiling for real */
5292
5293 else for (i = repeat_max - 1; i >= 0; i--)
5294 {
5295 pcre_uchar *hc;
5296 pcre_uchar *this_hwm = cd->hwm;
5297
5298 *code++ = OP_BRAZERO + repeat_type;
5299
5300 /* All but the final copy start a new nesting, maintaining the
5301 chain of brackets outstanding. */
5302
5303 if (i != 0)
5304 {
5305 int offset;
5306 *code++ = OP_BRA;
5307 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5308 bralink = code;
5309 PUTINC(code, 0, offset);
5310 }
5311
5312 memcpy(code, previous, IN_UCHARS(len));
5313
5314 /* Ensure there is enough workspace for forward references before
5315 copying them. */
5316
5317 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5318 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5319 {
5320 int save_offset = save_hwm - cd->start_workspace;
5321 int this_offset = this_hwm - cd->start_workspace;
5322 *errorcodeptr = expand_workspace(cd);
5323 if (*errorcodeptr != 0) goto FAILED;
5324 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5325 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5326 }
5327
5328 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5329 {
5330 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5331 cd->hwm += LINK_SIZE;
5332 }
5333 save_hwm = this_hwm;
5334 code += len;
5335 }
5336
5337 /* Now chain through the pending brackets, and fill in their length
5338 fields (which are holding the chain links pro tem). */
5339
5340 while (bralink != NULL)
5341 {
5342 int oldlinkoffset;
5343 int offset = (int)(code - bralink + 1);
5344 pcre_uchar *bra = code - offset;
5345 oldlinkoffset = GET(bra, 1);
5346 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5347 *code++ = OP_KET;
5348 PUTINC(code, 0, offset);
5349 PUT(bra, 1, offset);
5350 }
5351 }
5352
5353 /* If the maximum is unlimited, set a repeater in the final copy. For
5354 ONCE brackets, that's all we need to do. However, possessively repeated
5355 ONCE brackets can be converted into non-capturing brackets, as the
5356 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5357 deal with possessive ONCEs specially.
5358
5359 Otherwise, when we are doing the actual compile phase, check to see
5360 whether this group is one that could match an empty string. If so,
5361 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5362 that runtime checking can be done. [This check is also applied to ONCE
5363 groups at runtime, but in a different way.]
5364
5365 Then, if the quantifier was possessive and the bracket is not a
5366 conditional, we convert the BRA code to the POS form, and the KET code to
5367 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5368 subpattern at both the start and at the end.) The use of special opcodes
5369 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5370 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5371
5372 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5373 flag so that the default action below, of wrapping everything inside
5374 atomic brackets, does not happen. When the minimum is greater than 1,
5375 there will be earlier copies of the group, and so we still have to wrap
5376 the whole thing. */
5377
5378 else
5379 {
5380 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5381 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5382
5383 /* Convert possessive ONCE brackets to non-capturing */
5384
5385 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5386 possessive_quantifier) *bracode = OP_BRA;
5387
5388 /* For non-possessive ONCE brackets, all we need to do is to
5389 set the KET. */
5390
5391 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5392 *ketcode = OP_KETRMAX + repeat_type;
5393
5394 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5395 converted to non-capturing above). */
5396
5397 else
5398 {
5399 /* In the compile phase, check for empty string matching. */
5400
5401 if (lengthptr == NULL)
5402 {
5403 pcre_uchar *scode = bracode;
5404 do
5405 {
5406 if (could_be_empty_branch(scode, ketcode, utf, cd))
5407 {
5408 *bracode += OP_SBRA - OP_BRA;
5409 break;
5410 }
5411 scode += GET(scode, 1);
5412 }
5413 while (*scode == OP_ALT);
5414 }
5415
5416 /* Handle possessive quantifiers. */
5417
5418 if (possessive_quantifier)
5419 {
5420 /* For COND brackets, we wrap the whole thing in a possessively
5421 repeated non-capturing bracket, because we have not invented POS
5422 versions of the COND opcodes. Because we are moving code along, we
5423 must ensure that any pending recursive references are updated. */
5424
5425 if (*bracode == OP_COND || *bracode == OP_SCOND)
5426 {
5427 int nlen = (int)(code - bracode);
5428 *code = OP_END;
5429 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5430 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5431 code += 1 + LINK_SIZE;
5432 nlen += 1 + LINK_SIZE;
5433 *bracode = OP_BRAPOS;
5434 *code++ = OP_KETRPOS;
5435 PUTINC(code, 0, nlen);
5436 PUT(bracode, 1, nlen);
5437 }
5438
5439 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5440
5441 else
5442 {
5443 *bracode += 1; /* Switch to xxxPOS opcodes */
5444 *ketcode = OP_KETRPOS;
5445 }
5446
5447 /* If the minimum is zero, mark it as possessive, then unset the
5448 possessive flag when the minimum is 0 or 1. */
5449
5450 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5451 if (repeat_min < 2) possessive_quantifier = FALSE;
5452 }
5453
5454 /* Non-possessive quantifier */
5455
5456 else *ketcode = OP_KETRMAX + repeat_type;
5457 }
5458 }
5459 }
5460
5461 /* If previous is OP_FAIL, it was generated by an empty class [] in
5462 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5463 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5464 error above. We can just ignore the repeat in JS case. */
5465
5466 else if (*previous == OP_FAIL) goto END_REPEAT;
5467
5468 /* Else there's some kind of shambles */
5469
5470 else
5471 {
5472 *errorcodeptr = ERR11;
5473 goto FAILED;
5474 }
5475
5476 /* If the character following a repeat is '+', or if certain optimization
5477 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5478 there are special alternative opcodes for this case. For anything else, we
5479 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5480 notation is just syntactic sugar, taken from Sun's Java package, but the
5481 special opcodes can optimize it.
5482
5483 Some (but not all) possessively repeated subpatterns have already been
5484 completely handled in the code just above. For them, possessive_quantifier
5485 is always FALSE at this stage.
5486
5487 Note that the repeated item starts at tempcode, not at previous, which
5488 might be the first part of a string whose (former) last char we repeated.
5489
5490 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5491 an 'upto' may follow. We skip over an 'exact' item, and then test the
5492 length of what remains before proceeding. */
5493
5494 if (possessive_quantifier)
5495 {
5496 int len;
5497
5498 if (*tempcode == OP_TYPEEXACT)
5499 tempcode += PRIV(OP_lengths)[*tempcode] +
5500 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5501 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5502
5503 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5504 {
5505 tempcode += PRIV(OP_lengths)[*tempcode];
5506 #ifdef SUPPORT_UTF
5507 if (utf && HAS_EXTRALEN(tempcode[-1]))
5508 tempcode += GET_EXTRALEN(tempcode[-1]);
5509 #endif
5510 }
5511
5512 len = (int)(code - tempcode);
5513 if (len > 0) switch (*tempcode)
5514 {
5515 case OP_STAR: *tempcode = OP_POSSTAR; break;
5516 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5517 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5518 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5519
5520 case OP_STARI: *tempcode = OP_POSSTARI; break;
5521 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5522 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5523 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5524
5525 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5526 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5527 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5528 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5529
5530 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5531 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5532 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5533 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5534
5535 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5536 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5537 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5538 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5539
5540 /* Because we are moving code along, we must ensure that any
5541 pending recursive references are updated. */
5542
5543 default:
5544 *code = OP_END;
5545 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5546 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5547 code += 1 + LINK_SIZE;
5548 len += 1 + LINK_SIZE;
5549 tempcode[0] = OP_ONCE;
5550 *code++ = OP_KET;
5551 PUTINC(code, 0, len);
5552 PUT(tempcode, 1, len);
5553 break;
5554 }
5555 }
5556
5557 /* In all case we no longer have a previous item. We also set the
5558 "follows varying string" flag for subsequently encountered reqchars if
5559 it isn't already set and we have just passed a varying length item. */
5560
5561 END_REPEAT:
5562 previous = NULL;
5563 cd->req_varyopt |= reqvary;
5564 break;
5565
5566
5567 /* ===================================================================*/
5568 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5569 lookbehind or option setting or condition or all the other extended
5570 parenthesis forms. */
5571
5572 case CHAR_LEFT_PARENTHESIS:
5573 newoptions = options;
5574 skipbytes = 0;
5575 bravalue = OP_CBRA;
5576 save_hwm = cd->hwm;
5577 reset_bracount = FALSE;
5578
5579 /* First deal with various "verbs" that can be introduced by '*'. */
5580
5581 ptr++;
5582 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5583 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5584 {
5585 int i, namelen;
5586 int arglen = 0;
5587 const char *vn = verbnames;
5588 const pcre_uchar *name = ptr + 1;
5589 const pcre_uchar *arg = NULL;
5590 previous = NULL;
5591 ptr++;
5592 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5593 namelen = (int)(ptr - name);
5594
5595 /* It appears that Perl allows any characters whatsoever, other than
5596 a closing parenthesis, to appear in arguments, so we no longer insist on
5597 letters, digits, and underscores. */
5598
5599 if (*ptr == CHAR_COLON)
5600 {
5601 arg = ++ptr;
5602 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5603 arglen = (int)(ptr - arg);
5604 }
5605
5606 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5607 {
5608 *errorcodeptr = ERR60;
5609 goto FAILED;
5610 }
5611
5612 /* Scan the table of verb names */
5613
5614 for (i = 0; i < verbcount; i++)
5615 {
5616 if (namelen == verbs[i].len &&
5617 STRNCMP_UC_C8(name, vn, namelen) == 0)
5618 {
5619 /* Check for open captures before ACCEPT and convert it to
5620 ASSERT_ACCEPT if in an assertion. */
5621
5622 if (verbs[i].op == OP_ACCEPT)
5623 {
5624 open_capitem *oc;
5625 if (arglen != 0)
5626 {
5627 *errorcodeptr = ERR59;
5628 goto FAILED;
5629 }
5630 cd->had_accept = TRUE;
5631 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5632 {
5633 *code++ = OP_CLOSE;
5634 PUT2INC(code, 0, oc->number);
5635 }
5636 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5637
5638 /* Do not set firstchar after *ACCEPT */
5639 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5640 }
5641
5642 /* Handle other cases with/without an argument */
5643
5644 else if (arglen == 0)
5645 {
5646 if (verbs[i].op < 0) /* Argument is mandatory */
5647 {
5648 *errorcodeptr = ERR66;
5649 goto FAILED;
5650 }
5651 *code = verbs[i].op;
5652 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5653 }
5654
5655 else
5656 {
5657 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5658 {
5659 *errorcodeptr = ERR59;
5660 goto FAILED;
5661 }
5662 *code = verbs[i].op_arg;
5663 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5664 *code++ = arglen;
5665 memcpy(code, arg, IN_UCHARS(arglen));
5666 code += arglen;
5667 *code++ = 0;
5668 }
5669
5670 break; /* Found verb, exit loop */
5671 }
5672
5673 vn += verbs[i].len + 1;
5674 }
5675
5676 if (i < verbcount) continue; /* Successfully handled a verb */
5677 *errorcodeptr = ERR60; /* Verb not recognized */
5678 goto FAILED;
5679 }
5680
5681 /* Deal with the extended parentheses; all are introduced by '?', and the
5682 appearance of any of them means that this is not a capturing group. */
5683
5684 else if (*ptr == CHAR_QUESTION_MARK)
5685 {
5686 int i, set, unset, namelen;
5687 int *optset;
5688 const pcre_uchar *name;
5689 pcre_uchar *slot;
5690
5691 switch (*(++ptr))
5692 {
5693 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5694 ptr++;
5695 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5696 if (*ptr == 0)
5697 {
5698 *errorcodeptr = ERR18;
5699 goto FAILED;
5700 }
5701 continue;
5702
5703
5704 /* ------------------------------------------------------------ */
5705 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5706 reset_bracount = TRUE;
5707 /* Fall through */
5708
5709 /* ------------------------------------------------------------ */
5710 case CHAR_COLON: /* Non-capturing bracket */
5711 bravalue = OP_BRA;
5712 ptr++;
5713 break;
5714
5715
5716 /* ------------------------------------------------------------ */
5717 case CHAR_LEFT_PARENTHESIS:
5718 bravalue = OP_COND; /* Conditional group */
5719
5720 /* A condition can be an assertion, a number (referring to a numbered
5721 group), a name (referring to a named group), or 'R', referring to
5722 recursion. R<digits> and R&name are also permitted for recursion tests.
5723
5724 There are several syntaxes for testing a named group: (?(name)) is used
5725 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5726
5727 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5728 be the recursive thing or the name 'R' (and similarly for 'R' followed
5729 by digits), and (b) a number could be a name that consists of digits.
5730 In both cases, we look for a name first; if not found, we try the other
5731 cases. */
5732
5733 /* For conditions that are assertions, check the syntax, and then exit
5734 the switch. This will take control down to where bracketed groups,
5735 including assertions, are processed. */
5736
5737 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5738 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5739 break;
5740
5741 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5742 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5743
5744 code[1+LINK_SIZE] = OP_CREF;
5745 skipbytes = 1+IMM2_SIZE;
5746 refsign = -1;
5747
5748 /* Check for a test for recursion in a named group. */
5749
5750 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5751 {
5752 terminator = -1;
5753 ptr += 2;
5754 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5755 }
5756
5757 /* Check for a test for a named group's having been set, using the Perl
5758 syntax (?(<name>) or (?('name') */
5759
5760 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5761 {
5762 terminator = CHAR_GREATER_THAN_SIGN;
5763 ptr++;
5764 }
5765 else if (ptr[1] == CHAR_APOSTROPHE)
5766 {
5767 terminator = CHAR_APOSTROPHE;
5768 ptr++;
5769 }
5770 else
5771 {
5772 terminator = 0;
5773 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5774 }
5775
5776 /* We now expect to read a name; any thing else is an error */
5777
5778 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5779 {
5780 ptr += 1; /* To get the right offset */
5781 *errorcodeptr = ERR28;
5782 goto FAILED;
5783 }
5784
5785 /* Read the name, but also get it as a number if it's all digits */
5786
5787 recno = 0;
5788 name = ++ptr;
5789 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5790 {
5791 if (recno >= 0)
5792 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5793 ptr++;
5794 }
5795 namelen = (int)(ptr - name);
5796
5797 if ((terminator > 0 && *ptr++ != terminator) ||
5798 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5799 {
5800 ptr--; /* Error offset */
5801 *errorcodeptr = ERR26;
5802 goto FAILED;
5803 }
5804
5805 /* Do no further checking in the pre-compile phase. */
5806
5807 if (lengthptr != NULL) break;
5808
5809 /* In the real compile we do the work of looking for the actual
5810 reference. If the string started with "+" or "-" we require the rest to
5811 be digits, in which case recno will be set. */
5812
5813 if (refsign > 0)
5814 {
5815 if (recno <= 0)
5816 {
5817 *errorcodeptr = ERR58;
5818 goto FAILED;
5819 }
5820 recno = (refsign == CHAR_MINUS)?
5821 cd->bracount - recno + 1 : recno +cd->bracount;
5822 if (recno <= 0 || recno > cd->final_bracount)
5823 {
5824 *errorcodeptr = ERR15;
5825 goto FAILED;
5826 }
5827 PUT2(code, 2+LINK_SIZE, recno);
5828 break;
5829 }
5830
5831 /* Otherwise (did not start with "+" or "-"), start by looking for the
5832 name. If we find a name, add one to the opcode to change OP_CREF or
5833 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5834 except they record that the reference was originally to a name. The
5835 information is used to check duplicate names. */
5836
5837 slot = cd->name_table;
5838 for (i = 0; i < cd->names_found; i++)
5839 {
5840 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5841 slot += cd->name_entry_size;
5842 }
5843
5844 /* Found a previous named subpattern */
5845
5846 if (i < cd->names_found)
5847 {
5848 recno = GET2(slot, 0);
5849 PUT2(code, 2+LINK_SIZE, recno);
5850 code[1+LINK_SIZE]++;
5851 }
5852
5853 /* Search the pattern for a forward reference */
5854
5855 else if ((i = find_parens(cd, name, namelen,
5856 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5857 {
5858 PUT2(code, 2+LINK_SIZE, i);
5859 code[1+LINK_SIZE]++;
5860 }
5861
5862 /* If terminator == 0 it means that the name followed directly after
5863 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5864 some further alternatives to try. For the cases where terminator != 0
5865 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5866 now checked all the possibilities, so give an error. */
5867
5868 else if (terminator != 0)
5869 {
5870 *errorcodeptr = ERR15;
5871 goto FAILED;
5872 }
5873
5874 /* Check for (?(R) for recursion. Allow digits after R to specify a
5875 specific group number. */
5876
5877 else if (*name == CHAR_R)
5878 {
5879 recno = 0;
5880 for (i = 1; i < namelen; i++)
5881 {
5882 if (!IS_DIGIT(name[i]))
5883 {
5884 *errorcodeptr = ERR15;
5885 goto FAILED;
5886 }
5887 recno = recno * 10 + name[i] - CHAR_0;
5888 }
5889 if (recno == 0) recno = RREF_ANY;
5890 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5891 PUT2(code, 2+LINK_SIZE, recno);
5892 }
5893
5894 /* Similarly, check for the (?(DEFINE) "condition", which is always
5895 false. */
5896
5897 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5898 {
5899 code[1+LINK_SIZE] = OP_DEF;
5900 skipbytes = 1;
5901 }
5902
5903 /* Check for the "name" actually being a subpattern number. We are
5904 in the second pass here, so final_bracount is set. */
5905
5906 else if (recno > 0 && recno <= cd->final_bracount)
5907 {
5908 PUT2(code, 2+LINK_SIZE, recno);
5909 }
5910
5911 /* Either an unidentified subpattern, or a reference to (?(0) */
5912
5913 else
5914 {
5915 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5916 goto FAILED;
5917 }
5918 break;
5919
5920
5921 /* ------------------------------------------------------------ */
5922 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5923 bravalue = OP_ASSERT;
5924 cd->assert_depth += 1;
5925 ptr++;
5926 break;
5927
5928
5929 /* ------------------------------------------------------------ */
5930 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5931 ptr++;
5932 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5933 {
5934 *code++ = OP_FAIL;
5935 previous = NULL;
5936 continue;
5937 }
5938 bravalue = OP_ASSERT_NOT;
5939 cd->assert_depth += 1;
5940 break;
5941
5942
5943 /* ------------------------------------------------------------ */
5944 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5945 switch (ptr[1])
5946 {
5947 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5948 bravalue = OP_ASSERTBACK;
5949 cd->assert_depth += 1;
5950 ptr += 2;
5951 break;
5952
5953 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5954 bravalue = OP_ASSERTBACK_NOT;
5955 cd->assert_depth += 1;
5956 ptr += 2;
5957 break;
5958
5959 default: /* Could be name define, else bad */
5960 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5961 goto DEFINE_NAME;
5962 ptr++; /* Correct offset for error */
5963 *errorcodeptr = ERR24;
5964 goto FAILED;
5965 }
5966 break;
5967
5968
5969 /* ------------------------------------------------------------ */
5970 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5971 bravalue = OP_ONCE;
5972 ptr++;
5973 break;
5974
5975
5976 /* ------------------------------------------------------------ */
5977 case CHAR_C: /* Callout - may be followed by digits; */
5978 previous_callout = code; /* Save for later completion */
5979 after_manual_callout = 1; /* Skip one item before completing */
5980 *code++ = OP_CALLOUT;
5981 {
5982 int n = 0;
5983 ptr++;
5984 while(IS_DIGIT(*ptr))
5985 n = n * 10 + *ptr++ - CHAR_0;
5986 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5987 {
5988 *errorcodeptr = ERR39;
5989 goto FAILED;
5990 }
5991 if (n > 255)
5992 {
5993 *errorcodeptr = ERR38;
5994 goto FAILED;
5995 }
5996 *code++ = n;
5997 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5998 PUT(code, LINK_SIZE, 0); /* Default length */
5999 code += 2 * LINK_SIZE;
6000 }
6001 previous = NULL;
6002 continue;
6003
6004
6005 /* ------------------------------------------------------------ */
6006 case CHAR_P: /* Python-style named subpattern handling */
6007 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6008 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6009 {
6010 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6011 terminator = CHAR_RIGHT_PARENTHESIS;
6012 goto NAMED_REF_OR_RECURSE;
6013 }
6014 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6015 {
6016 *errorcodeptr = ERR41;
6017 goto FAILED;
6018 }
6019 /* Fall through to handle (?P< as (?< is handled */
6020
6021
6022 /* ------------------------------------------------------------ */
6023 DEFINE_NAME: /* Come here from (?< handling */
6024 case CHAR_APOSTROPHE:
6025 {
6026 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6027 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6028 name = ++ptr;
6029
6030 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6031 namelen = (int)(ptr - name);
6032
6033 /* In the pre-compile phase, just do a syntax check. */
6034
6035 if (lengthptr != NULL)
6036 {
6037 if (*ptr != terminator)
6038 {
6039 *errorcodeptr = ERR42;
6040 goto FAILED;
6041 }
6042 if (cd->names_found >= MAX_NAME_COUNT)
6043 {
6044 *errorcodeptr = ERR49;
6045 goto FAILED;
6046 }
6047 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6048 {
6049 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6050 if (namelen > MAX_NAME_SIZE)
6051 {
6052 *errorcodeptr = ERR48;
6053 goto FAILED;
6054 }
6055 }
6056 }
6057
6058 /* In the real compile, create the entry in the table, maintaining
6059 alphabetical order. Duplicate names for different numbers are
6060 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6061 number are always OK. (An existing number can be re-used if (?|
6062 appears in the pattern.) In either event, a duplicate name results in
6063 a duplicate entry in the table, even if the number is the same. This
6064 is because the number of names, and hence the table size, is computed
6065 in the pre-compile, and it affects various numbers and pointers which
6066 would all have to be modified, and the compiled code moved down, if
6067 duplicates with the same number were omitted from the table. This
6068 doesn't seem worth the hassle. However, *different* names for the
6069 same number are not permitted. */
6070
6071 else
6072 {
6073 BOOL dupname = FALSE;
6074 slot = cd->name_table;
6075
6076 for (i = 0; i < cd->names_found; i++)
6077 {
6078 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6079 if (crc == 0)
6080 {
6081 if (slot[IMM2_SIZE+namelen] == 0)
6082 {
6083 if (GET2(slot, 0) != cd->bracount + 1 &&
6084 (options & PCRE_DUPNAMES) == 0)
6085 {
6086 *errorcodeptr = ERR43;
6087 goto FAILED;
6088 }
6089 else dupname = TRUE;
6090 }
6091 else crc = -1; /* Current name is a substring */
6092 }
6093
6094 /* Make space in the table and break the loop for an earlier
6095 name. For a duplicate or later name, carry on. We do this for
6096 duplicates so that in the simple case (when ?(| is not used) they
6097 are in order of their numbers. */
6098
6099 if (crc < 0)
6100 {
6101 memmove(slot + cd->name_entry_size, slot,
6102 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6103 break;
6104 }
6105
6106 /* Continue the loop for a later or duplicate name */
6107
6108 slot += cd->name_entry_size;
6109 }
6110
6111 /* For non-duplicate names, check for a duplicate number before
6112 adding the new name. */
6113
6114 if (!dupname)
6115 {
6116 pcre_uchar *cslot = cd->name_table;
6117 for (i = 0; i < cd->names_found; i++)
6118 {
6119 if (cslot != slot)
6120 {
6121 if (GET2(cslot, 0) == cd->bracount + 1)
6122 {
6123 *errorcodeptr = ERR65;
6124 goto FAILED;
6125 }
6126 }
6127 else i--;
6128 cslot += cd->name_entry_size;
6129 }
6130 }
6131
6132 PUT2(slot, 0, cd->bracount + 1);
6133 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6134 slot[IMM2_SIZE + namelen] = 0;
6135 }
6136 }
6137
6138 /* In both pre-compile and compile, count the number of names we've
6139 encountered. */
6140
6141 cd->names_found++;
6142 ptr++; /* Move past > or ' */
6143 goto NUMBERED_GROUP;
6144
6145
6146 /* ------------------------------------------------------------ */
6147 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6148 terminator = CHAR_RIGHT_PARENTHESIS;
6149 is_recurse = TRUE;
6150 /* Fall through */
6151
6152 /* We come here from the Python syntax above that handles both
6153 references (?P=name) and recursion (?P>name), as well as falling
6154 through from the Perl recursion syntax (?&name). We also come here from
6155 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6156 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6157
6158 NAMED_REF_OR_RECURSE:
6159 name = ++ptr;
6160 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6161 namelen = (int)(ptr - name);
6162
6163 /* In the pre-compile phase, do a syntax check. We used to just set
6164 a dummy reference number, because it was not used in the first pass.
6165 However, with the change of recursive back references to be atomic,
6166 we have to look for the number so that this state can be identified, as
6167 otherwise the incorrect length is computed. If it's not a backwards
6168 reference, the dummy number will do. */
6169
6170 if (lengthptr != NULL)
6171 {
6172 const pcre_uchar *temp;
6173
6174 if (namelen == 0)
6175 {
6176 *errorcodeptr = ERR62;
6177 goto FAILED;
6178 }
6179 if (*ptr != terminator)
6180 {
6181 *errorcodeptr = ERR42;
6182 goto FAILED;
6183 }
6184 if (namelen > MAX_NAME_SIZE)
6185 {
6186 *errorcodeptr = ERR48;
6187 goto FAILED;
6188 }
6189
6190 /* The name table does not exist in the first pass, so we cannot
6191 do a simple search as in the code below. Instead, we have to scan the
6192 pattern to find the number. It is important that we scan it only as
6193 far as we have got because the syntax of named subpatterns has not
6194 been checked for the rest of the pattern, and find_parens() assumes
6195 correct syntax. In any case, it's a waste of resources to scan
6196 further. We stop the scan at the current point by temporarily
6197 adjusting the value of cd->endpattern. */
6198
6199 temp = cd->end_pattern;
6200 cd->end_pattern = ptr;
6201 recno = find_parens(cd, name, namelen,
6202 (options & PCRE_EXTENDED) != 0, utf);
6203 cd->end_pattern = temp;
6204 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6205 }
6206
6207 /* In the real compile, seek the name in the table. We check the name
6208 first, and then check that we have reached the end of the name in the
6209 table. That way, if the name that is longer than any in the table,
6210 the comparison will fail without reading beyond the table entry. */
6211
6212 else
6213 {
6214 slot = cd->name_table;
6215 for (i = 0; i < cd->names_found; i++)
6216 {
6217 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6218 slot[IMM2_SIZE+namelen] == 0)
6219 break;
6220 slot += cd->name_entry_size;
6221 }
6222
6223 if (i < cd->names_found) /* Back reference */
6224 {
6225 recno = GET2(slot, 0);
6226 }
6227 else if ((recno = /* Forward back reference */
6228 find_parens(cd, name, namelen,
6229 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6230 {
6231 *errorcodeptr = ERR15;
6232 goto FAILED;
6233 }
6234 }
6235
6236 /* In both phases, we can now go to the code than handles numerical
6237 recursion or backreferences. */
6238
6239 if (is_recurse) goto HANDLE_RECURSION;
6240 else goto HANDLE_REFERENCE;
6241
6242
6243 /* ------------------------------------------------------------ */
6244 case CHAR_R: /* Recursion */
6245 ptr++; /* Same as (?0) */
6246 /* Fall through */
6247
6248
6249 /* ------------------------------------------------------------ */
6250 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6251 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6252 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6253 {
6254 const pcre_uchar *called;
6255 terminator = CHAR_RIGHT_PARENTHESIS;
6256
6257 /* Come here from the \g<...> and \g'...' code (Oniguruma
6258 compatibility). However, the syntax has been checked to ensure that
6259 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6260 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6261 ever be taken. */
6262
6263 HANDLE_NUMERICAL_RECURSION:
6264
6265 if ((refsign = *ptr) == CHAR_PLUS)
6266 {
6267 ptr++;
6268 if (!IS_DIGIT(*ptr))
6269 {
6270 *errorcodeptr = ERR63;
6271 goto FAILED;
6272 }
6273 }
6274 else if (refsign == CHAR_MINUS)
6275 {
6276 if (!IS_DIGIT(ptr[1]))
6277 goto OTHER_CHAR_AFTER_QUERY;
6278 ptr++;
6279 }
6280
6281 recno = 0;
6282 while(IS_DIGIT(*ptr))
6283 recno = recno * 10 + *ptr++ - CHAR_0;
6284
6285 if (*ptr != terminator)
6286 {
6287 *errorcodeptr = ERR29;
6288 goto FAILED;
6289 }
6290
6291 if (refsign == CHAR_MINUS)
6292 {
6293 if (recno == 0)
6294 {
6295 *errorcodeptr = ERR58;
6296 goto FAILED;
6297 }
6298 recno = cd->bracount - recno + 1;
6299 if (recno <= 0)
6300 {
6301 *errorcodeptr = ERR15;
6302 goto FAILED;
6303 }
6304 }
6305 else if (refsign == CHAR_PLUS)
6306 {
6307 if (recno == 0)
6308 {
6309 *errorcodeptr = ERR58;
6310 goto FAILED;
6311 }
6312 recno += cd->bracount;
6313 }
6314
6315 /* Come here from code above that handles a named recursion */
6316
6317 HANDLE_RECURSION:
6318
6319 previous = code;
6320 called = cd->start_code;
6321
6322 /* When we are actually compiling, find the bracket that is being
6323 referenced. Temporarily end the regex in case it doesn't exist before
6324 this point. If we end up with a forward reference, first check that
6325 the bracket does occur later so we can give the error (and position)
6326 now. Then remember this forward reference in the workspace so it can
6327 be filled in at the end. */
6328
6329 if (lengthptr == NULL)
6330 {
6331 *code = OP_END;
6332 if (recno != 0)
6333 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6334
6335 /* Forward reference */
6336
6337 if (called == NULL)
6338 {
6339 if (find_parens(cd, NULL, recno,
6340 (options & PCRE_EXTENDED) != 0, utf) < 0)
6341 {
6342 *errorcodeptr = ERR15;
6343 goto FAILED;
6344 }
6345
6346 /* Fudge the value of "called" so that when it is inserted as an
6347 offset below, what it actually inserted is the reference number
6348 of the group. Then remember the forward reference. */
6349
6350 called = cd->start_code + recno;
6351 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6352 WORK_SIZE_SAFETY_MARGIN)
6353 {
6354 *errorcodeptr = expand_workspace(cd);
6355 if (*errorcodeptr != 0) goto FAILED;
6356 }
6357 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6358 }
6359
6360 /* If not a forward reference, and the subpattern is still open,
6361 this is a recursive call. We check to see if this is a left
6362 recursion that could loop for ever, and diagnose that case. We
6363 must not, however, do this check if we are in a conditional
6364 subpattern because the condition might be testing for recursion in
6365 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6366 Forever loops are also detected at runtime, so those that occur in
6367 conditional subpatterns will be picked up then. */
6368
6369 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6370 could_be_empty(called, code, bcptr, utf, cd))
6371 {
6372 *errorcodeptr = ERR40;
6373 goto FAILED;
6374 }
6375 }
6376
6377 /* Insert the recursion/subroutine item. It does not have a set first
6378 character (relevant if it is repeated, because it will then be
6379 wrapped with ONCE brackets). */
6380
6381 *code = OP_RECURSE;
6382 PUT(code, 1, (int)(called - cd->start_code));
6383 code += 1 + LINK_SIZE;
6384 groupsetfirstchar = FALSE;
6385 }
6386
6387 /* Can't determine a first byte now */
6388
6389 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6390 continue;
6391
6392
6393 /* ------------------------------------------------------------ */
6394 default: /* Other characters: check option setting */
6395 OTHER_CHAR_AFTER_QUERY:
6396 set = unset = 0;
6397 optset = &set;
6398
6399 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6400 {
6401 switch (*ptr++)
6402 {
6403 case CHAR_MINUS: optset = &unset; break;
6404
6405 case CHAR_J: /* Record that it changed in the external options */
6406 *optset |= PCRE_DUPNAMES;
6407 cd->external_flags |= PCRE_JCHANGED;
6408 break;
6409
6410 case CHAR_i: *optset |= PCRE_CASELESS; break;
6411 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6412 case CHAR_s: *optset |= PCRE_DOTALL; break;
6413 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6414 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6415 case CHAR_X: *optset |= PCRE_EXTRA; break;
6416
6417 default: *errorcodeptr = ERR12;
6418 ptr--; /* Correct the offset */
6419 goto FAILED;
6420 }
6421 }
6422
6423 /* Set up the changed option bits, but don't change anything yet. */
6424
6425 newoptions = (options | set) & (~unset);
6426
6427 /* If the options ended with ')' this is not the start of a nested
6428 group with option changes, so the options change at this level. If this
6429 item is right at the start of the pattern, the options can be
6430 abstracted and made external in the pre-compile phase, and ignored in
6431 the compile phase. This can be helpful when matching -- for instance in
6432 caseless checking of required bytes.
6433
6434 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6435 definitely *not* at the start of the pattern because something has been
6436 compiled. In the pre-compile phase, however, the code pointer can have
6437 that value after the start, because it gets reset as code is discarded
6438 during the pre-compile. However, this can happen only at top level - if
6439 we are within parentheses, the starting BRA will still be present. At
6440 any parenthesis level, the length value can be used to test if anything
6441 has been compiled at that level. Thus, a test for both these conditions
6442 is necessary to ensure we correctly detect the start of the pattern in
6443 both phases.
6444
6445 If we are not at the pattern start, reset the greedy defaults and the
6446 case value for firstchar and reqchar. */
6447
6448 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6449 {
6450 if (code == cd->start_code + 1 + LINK_SIZE &&
6451 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6452 {
6453 cd->external_options = newoptions;
6454 }
6455 else
6456 {
6457 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6458 greedy_non_default = greedy_default ^ 1;
6459 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6460 }
6461
6462 /* Change options at this level, and pass them back for use
6463 in subsequent branches. */
6464
6465 *optionsptr = options = newoptions;
6466 previous = NULL; /* This item can't be repeated */
6467 continue; /* It is complete */
6468 }
6469
6470 /* If the options ended with ':' we are heading into a nested group
6471 with possible change of options. Such groups are non-capturing and are
6472 not assertions of any kind. All we need to do is skip over the ':';
6473 the newoptions value is handled below. */
6474
6475 bravalue = OP_BRA;
6476 ptr++;
6477 } /* End of switch for character following (? */
6478 } /* End of (? handling */
6479
6480 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6481 is set, all unadorned brackets become non-capturing and behave like (?:...)
6482 brackets. */
6483
6484 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6485 {
6486 bravalue = OP_BRA;
6487 }
6488
6489 /* Else we have a capturing group. */
6490
6491 else
6492 {
6493 NUMBERED_GROUP:
6494 cd->bracount += 1;
6495 PUT2(code, 1+LINK_SIZE, cd->bracount);
6496 skipbytes = IMM2_SIZE;
6497 }
6498
6499 /* Process nested bracketed regex. Assertions used not to be repeatable,
6500 but this was changed for Perl compatibility, so all kinds can now be
6501 repeated. We copy code into a non-register variable (tempcode) in order to
6502 be able to pass its address because some compilers complain otherwise. */
6503
6504 previous = code; /* For handling repetition */
6505 *code = bravalue;
6506 tempcode = code;
6507 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6508 tempbracount = cd->bracount; /* Save value before bracket */
6509 length_prevgroup = 0; /* Initialize for pre-compile phase */
6510
6511 if (!compile_regex(
6512 newoptions, /* The complete new option state */
6513 &tempcode, /* Where to put code (updated) */
6514 &ptr, /* Input pointer (updated) */
6515 errorcodeptr, /* Where to put an error message */
6516 (bravalue == OP_ASSERTBACK ||
6517 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6518 reset_bracount, /* True if (?| group */
6519 skipbytes, /* Skip over bracket number */
6520 cond_depth +
6521 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6522 &subfirstchar, /* For possible first char */
6523 &subreqchar, /* For possible last char */
6524 bcptr, /* Current branch chain */
6525 cd, /* Tables block */
6526 (lengthptr == NULL)? NULL : /* Actual compile phase */
6527 &length_prevgroup /* Pre-compile phase */
6528 ))
6529 goto FAILED;
6530
6531 /* If this was an atomic group and there are no capturing groups within it,
6532 generate OP_ONCE_NC instead of OP_ONCE. */
6533
6534 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6535 *code = OP_ONCE_NC;
6536
6537 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6538 cd->assert_depth -= 1;
6539
6540 /* At the end of compiling, code is still pointing to the start of the
6541 group, while tempcode has been updated to point past the end of the group.
6542 The pattern pointer (ptr) is on the bracket.
6543
6544 If this is a conditional bracket, check that there are no more than
6545 two branches in the group, or just one if it's a DEFINE group. We do this
6546 in the real compile phase, not in the pre-pass, where the whole group may
6547 not be available. */
6548
6549 if (bravalue == OP_COND && lengthptr == NULL)
6550 {
6551 pcre_uchar *tc = code;
6552 int condcount = 0;
6553
6554 do {
6555 condcount++;
6556 tc += GET(tc,1);
6557 }
6558 while (*tc != OP_KET);
6559
6560 /* A DEFINE group is never obeyed inline (the "condition" is always
6561 false). It must have only one branch. */
6562
6563 if (code[LINK_SIZE+1] == OP_DEF)
6564 {
6565 if (condcount > 1)
6566 {
6567 *errorcodeptr = ERR54;
6568 goto FAILED;
6569 }
6570 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6571 }
6572
6573 /* A "normal" conditional group. If there is just one branch, we must not
6574 make use of its firstchar or reqchar, because this is equivalent to an
6575 empty second branch. */
6576
6577 else
6578 {
6579 if (condcount > 2)
6580 {
6581 *errorcodeptr = ERR27;
6582 goto FAILED;
6583 }
6584 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6585 }
6586 }
6587
6588 /* Error if hit end of pattern */
6589
6590 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6591 {
6592 *errorcodeptr = ERR14;
6593 goto FAILED;
6594 }
6595
6596 /* In the pre-compile phase, update the length by the length of the group,
6597 less the brackets at either end. Then reduce the compiled code to just a
6598 set of non-capturing brackets so that it doesn't use much memory if it is
6599 duplicated by a quantifier.*/
6600
6601 if (lengthptr != NULL)
6602 {
6603 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6604 {
6605 *errorcodeptr = ERR20;
6606 goto FAILED;
6607 }
6608 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6609 code++; /* This already contains bravalue */
6610 PUTINC(code, 0, 1 + LINK_SIZE);
6611 *code++ = OP_KET;
6612 PUTINC(code, 0, 1 + LINK_SIZE);
6613 break; /* No need to waste time with special character handling */
6614 }
6615
6616 /* Otherwise update the main code pointer to the end of the group. */
6617
6618 code = tempcode;
6619
6620 /* For a DEFINE group, required and first character settings are not
6621 relevant. */
6622
6623 if (bravalue == OP_DEF) break;
6624
6625 /* Handle updating of the required and first characters for other types of
6626 group. Update for normal brackets of all kinds, and conditions with two
6627 branches (see code above). If the bracket is followed by a quantifier with
6628 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6629 zerofirstchar outside the main loop so that they can be accessed for the
6630 back off. */
6631
6632 zeroreqchar = reqchar;
6633 zerofirstchar = firstchar;
6634 groupsetfirstchar = FALSE;
6635
6636 if (bravalue >= OP_ONCE)
6637 {
6638 /* If we have not yet set a firstchar in this branch, take it from the
6639 subpattern, remembering that it was set here so that a repeat of more
6640 than one can replicate it as reqchar if necessar